From e08f7a2c1c265a985b627357a096d58778ac57eb Mon Sep 17 00:00:00 2001 From: GamerBoss101 Date: Sat, 29 Mar 2025 22:27:44 -0400 Subject: [PATCH 01/10] Server API and Webpage update --- Backend/server.py | 127 ++++++++++++++++++++++++++++++++++++++++++++-- 1 file changed, 123 insertions(+), 4 deletions(-) diff --git a/Backend/server.py b/Backend/server.py index b9736b5..97b346b 100644 --- a/Backend/server.py +++ b/Backend/server.py @@ -9,7 +9,9 @@ import io import whisperx from io import BytesIO from typing import List, Dict, Any, Optional -from fastapi import FastAPI, WebSocket, WebSocketDisconnect +from fastapi import FastAPI, WebSocket, WebSocketDisconnect, Request +from fastapi.responses import HTMLResponse, FileResponse +from fastapi.staticfiles import StaticFiles from fastapi.middleware.cors import CORSMiddleware from pydantic import BaseModel from generator import load_csm_1b, Segment @@ -17,6 +19,8 @@ import uvicorn import time import gc from collections import deque +import socket +import requests # Select device if torch.cuda.is_available(): @@ -45,6 +49,32 @@ app.add_middleware( allow_headers=["*"], ) +# Define the base directory +base_dir = os.path.dirname(os.path.abspath(__file__)) + +# Mount a static files directory if you have any static assets like CSS or JS +static_dir = os.path.join(base_dir, "static") +os.makedirs(static_dir, exist_ok=True) # Create the directory if it doesn't exist +app.mount("/static", StaticFiles(directory=static_dir), name="static") + +# Define route to serve index.html as the main page +@app.get("/", response_class=HTMLResponse) +async def get_index(): + try: + with open(os.path.join(base_dir, "index.html"), "r") as f: + return HTMLResponse(content=f.read()) + except FileNotFoundError: + return HTMLResponse(content="

Error: index.html not found

") + +# Add a favicon endpoint (optional, but good to have) +@app.get("/favicon.ico") +async def get_favicon(): + favicon_path = os.path.join(static_dir, "favicon.ico") + if os.path.exists(favicon_path): + return FileResponse(favicon_path) + else: + return HTMLResponse(status_code=204) # No content + # Connection manager to handle multiple clients class ConnectionManager: def __init__(self): @@ -259,6 +289,7 @@ async def websocket_endpoint(websocket: WebSocket): energy_window.clear() is_silence = False last_active_time = time.time() + print(f"Streaming started with speaker ID: {speaker_id}") await websocket.send_json({ "type": "streaming_status", "status": "started" @@ -269,6 +300,13 @@ async def websocket_endpoint(websocket: WebSocket): energy_window.append(chunk_energy) avg_energy = sum(energy_window) / len(energy_window) + # Debug audio levels + if len(energy_window) >= 5: # Only start printing after we have enough samples + if avg_energy > SILENCE_THRESHOLD: + print(f"[AUDIO] Active sound detected - Energy: {avg_energy:.6f} (threshold: {SILENCE_THRESHOLD})") + else: + print(f"[AUDIO] Silence detected - Energy: {avg_energy:.6f} (threshold: {SILENCE_THRESHOLD})") + # Check if audio is silent current_silence = avg_energy < SILENCE_THRESHOLD @@ -277,33 +315,53 @@ async def websocket_endpoint(websocket: WebSocket): # Transition to silence is_silence = True last_active_time = time.time() + print("[STREAM] Transition to silence detected") elif is_silence and not current_silence: # User started talking again is_silence = False + print("[STREAM] User resumed speaking") # Add chunk to buffer regardless of silence state streaming_buffer.append(audio_chunk) + # Debug buffer size periodically + if len(streaming_buffer) % 10 == 0: + print(f"[BUFFER] Current size: {len(streaming_buffer)} chunks, ~{len(streaming_buffer)/5:.1f} seconds") + # Check if silence has persisted long enough to consider "stopped talking" silence_elapsed = time.time() - last_active_time if is_silence and silence_elapsed >= SILENCE_DURATION_SEC and len(streaming_buffer) > 0: # User has stopped talking - process the collected audio + print(f"[STREAM] Processing audio after {silence_elapsed:.2f}s of silence") + print(f"[STREAM] Processing {len(streaming_buffer)} audio chunks (~{len(streaming_buffer)/5:.1f} seconds)") + full_audio = torch.cat(streaming_buffer, dim=0) + # Log audio statistics + audio_duration = len(full_audio) / generator.sample_rate + audio_min = torch.min(full_audio).item() + audio_max = torch.max(full_audio).item() + audio_mean = torch.mean(full_audio).item() + print(f"[AUDIO] Processed audio - Duration: {audio_duration:.2f}s, Min: {audio_min:.4f}, Max: {audio_max:.4f}, Mean: {audio_mean:.4f}") + # Process with WhisperX speech-to-text + print("[ASR] Starting transcription with WhisperX...") transcribed_text = await transcribe_audio(full_audio) # Log the transcription - print(f"Transcribed text: '{transcribed_text}'") + print(f"[ASR] Transcribed text: '{transcribed_text}'") # Add to conversation context if transcribed_text: + print(f"[DIALOG] Adding user utterance to context: '{transcribed_text}'") user_segment = Segment(text=transcribed_text, speaker=speaker_id, audio=full_audio) context_segments.append(user_segment) # Generate a contextual response + print("[DIALOG] Generating response...") response_text = await generate_response(transcribed_text, context_segments) + print(f"[DIALOG] Response text: '{response_text}'") # Send the transcribed text to client await websocket.send_json({ @@ -312,12 +370,14 @@ async def websocket_endpoint(websocket: WebSocket): }) # Generate audio for the response + print("[TTS] Generating speech for response...") audio_tensor = generator.generate( text=response_text, speaker=1 if speaker_id == 0 else 0, # Use opposite speaker context=context_segments, max_audio_length_ms=10_000, ) + print(f"[TTS] Generated audio length: {len(audio_tensor)/generator.sample_rate:.2f}s") # Add response to context ai_segment = Segment( @@ -326,15 +386,18 @@ async def websocket_endpoint(websocket: WebSocket): audio=audio_tensor ) context_segments.append(ai_segment) + print(f"[DIALOG] Context now has {len(context_segments)} segments") # Convert audio to base64 and send back to client audio_base64 = await encode_audio_data(audio_tensor) + print("[STREAM] Sending audio response to client") await websocket.send_json({ "type": "audio_response", "text": response_text, "audio": audio_base64 }) else: + print("[ASR] Transcription failed or returned empty text") # If transcription failed, send a generic response await websocket.send_json({ "type": "error", @@ -346,17 +409,20 @@ async def websocket_endpoint(websocket: WebSocket): energy_window.clear() is_silence = False last_active_time = time.time() + print("[STREAM] Buffer cleared, ready for next utterance") # If buffer gets too large without silence, process it anyway # This prevents memory issues with very long streams elif len(streaming_buffer) >= 30: # ~6 seconds of audio at 5 chunks/sec - print("Buffer limit reached, processing audio") + print("[BUFFER] Maximum buffer size reached, processing audio") full_audio = torch.cat(streaming_buffer, dim=0) # Process with WhisperX speech-to-text + print("[ASR] Starting forced transcription of long audio...") transcribed_text = await transcribe_audio(full_audio) if transcribed_text: + print(f"[ASR] Transcribed long audio: '{transcribed_text}'") context_segments.append(Segment(text=transcribed_text, speaker=speaker_id, audio=full_audio)) # Send the transcribed text to client @@ -364,11 +430,17 @@ async def websocket_endpoint(websocket: WebSocket): "type": "transcription", "text": transcribed_text + " (processing continued speech...)" }) + else: + print("[ASR] No transcription from long audio") streaming_buffer = [] + print("[BUFFER] Buffer cleared due to size limit") except Exception as e: - print(f"Error processing streaming audio: {str(e)}") + print(f"[ERROR] Processing streaming audio: {str(e)}") + # Print traceback for more detailed error information + import traceback + traceback.print_exc() await websocket.send_json({ "type": "error", "message": f"Error processing streaming audio: {str(e)}" @@ -412,6 +484,53 @@ async def websocket_endpoint(websocket: WebSocket): pass manager.disconnect(websocket) +# Add this function to get the public IP address +def get_public_ip(): + """Get the server's public IP address using an external service""" + try: + # Try multiple services in case one is down + services = [ + "https://api.ipify.org", + "https://ifconfig.me/ip", + "https://checkip.amazonaws.com", + ] + + for service in services: + try: + response = requests.get(service, timeout=3) + if response.status_code == 200: + return response.text.strip() + except: + continue + + # Fallback to socket if external services fail + s = socket.socket(socket.AF_INET, socket.SOCK_DGRAM) + try: + # Doesn't need to be reachable, just used to determine interface + s.connect(('8.8.8.8', 1)) + local_ip = s.getsockname()[0] + return local_ip + except: + return "localhost" + finally: + s.close() + except: + return "Could not determine IP address" +# Update the __main__ block if __name__ == "__main__": + public_ip = get_public_ip() + print(f"\n{'='*50}") + print(f"💬 Sesame AI Voice Chat Server") + print(f"{'='*50}") + print(f"📡 Server Information:") + print(f" - Public IP: http://{public_ip}:8000") + print(f" - Local URL: http://localhost:8000") + print(f" - WebSocket: ws://{public_ip}:8000/ws") + print(f"{'='*50}") + print(f"🌐 Connect from web browsers using: http://{public_ip}:8000") + print(f"🔧 Serving index.html from: {os.path.join(base_dir, 'index.html')}") + print(f"{'='*50}\n") + + # Start the server uvicorn.run(app, host="0.0.0.0", port=8000) \ No newline at end of file From e1f976eaca156ed00d2659f61bfc93018128aaba Mon Sep 17 00:00:00 2001 From: GamerBoss101 Date: Sat, 29 Mar 2025 22:37:27 -0400 Subject: [PATCH 02/10] Server info print update --- Backend/server.py | 63 +++++++++++++---------------------------------- 1 file changed, 17 insertions(+), 46 deletions(-) diff --git a/Backend/server.py b/Backend/server.py index 97b346b..f159025 100644 --- a/Backend/server.py +++ b/Backend/server.py @@ -19,8 +19,6 @@ import uvicorn import time import gc from collections import deque -import socket -import requests # Select device if torch.cuda.is_available(): @@ -484,53 +482,26 @@ async def websocket_endpoint(websocket: WebSocket): pass manager.disconnect(websocket) -# Add this function to get the public IP address -def get_public_ip(): - """Get the server's public IP address using an external service""" - try: - # Try multiple services in case one is down - services = [ - "https://api.ipify.org", - "https://ifconfig.me/ip", - "https://checkip.amazonaws.com", - ] - - for service in services: - try: - response = requests.get(service, timeout=3) - if response.status_code == 200: - return response.text.strip() - except: - continue - - # Fallback to socket if external services fail - s = socket.socket(socket.AF_INET, socket.SOCK_DGRAM) - try: - # Doesn't need to be reachable, just used to determine interface - s.connect(('8.8.8.8', 1)) - local_ip = s.getsockname()[0] - return local_ip - except: - return "localhost" - finally: - s.close() - except: - return "Could not determine IP address" - -# Update the __main__ block +# Update the __main__ block with a comprehensive server startup message if __name__ == "__main__": - public_ip = get_public_ip() - print(f"\n{'='*50}") - print(f"💬 Sesame AI Voice Chat Server") - print(f"{'='*50}") + print(f"\n{'='*60}") + print(f"🔊 Sesame AI Voice Chat Server") + print(f"{'='*60}") print(f"📡 Server Information:") - print(f" - Public IP: http://{public_ip}:8000") print(f" - Local URL: http://localhost:8000") - print(f" - WebSocket: ws://{public_ip}:8000/ws") - print(f"{'='*50}") - print(f"🌐 Connect from web browsers using: http://{public_ip}:8000") - print(f"🔧 Serving index.html from: {os.path.join(base_dir, 'index.html')}") - print(f"{'='*50}\n") + print(f" - Network URL: http://:8000") + print(f" - WebSocket: ws://:8000/ws") + print(f"{'='*60}") + print(f"💡 To make this server public:") + print(f" 1. Ensure port 8000 is open in your firewall") + print(f" 2. Set up port forwarding on your router to port 8000") + print(f" 3. Or use a service like ngrok with: ngrok http 8000") + print(f"{'='*60}") + print(f"🌐 Device: {device.upper()}") + print(f"🧠 Models loaded: Sesame CSM + WhisperX ({asr_model.device})") + print(f"🔧 Serving from: {os.path.join(base_dir, 'index.html')}") + print(f"{'='*60}") + print(f"Ready to receive connections! Press Ctrl+C to stop the server.\n") # Start the server uvicorn.run(app, host="0.0.0.0", port=8000) \ No newline at end of file From 08fec9c403695598749fc27f6816a3351728230f Mon Sep 17 00:00:00 2001 From: GamerBoss101 Date: Sat, 29 Mar 2025 22:48:24 -0400 Subject: [PATCH 03/10] Server and Client Side update --- Backend/index.html | 555 ++++++++++++++++++++--------------- Backend/server.py | 707 +++++++++++++++++++++------------------------ 2 files changed, 655 insertions(+), 607 deletions(-) diff --git a/Backend/index.html b/Backend/index.html index 0e4006e..2944700 100644 --- a/Backend/index.html +++ b/Backend/index.html @@ -1,9 +1,13 @@ +/Backend/index.html --> Sesame AI Voice Chat + + + @@ -162,8 +204,8 @@ - - + +
@@ -173,7 +215,7 @@ \ No newline at end of file diff --git a/Backend/server.py b/Backend/server.py index f159025..e986606 100644 --- a/Backend/server.py +++ b/Backend/server.py @@ -1,24 +1,20 @@ import os import base64 import json -import asyncio import torch import torchaudio import numpy as np -import io import whisperx from io import BytesIO from typing import List, Dict, Any, Optional -from fastapi import FastAPI, WebSocket, WebSocketDisconnect, Request -from fastapi.responses import HTMLResponse, FileResponse -from fastapi.staticfiles import StaticFiles -from fastapi.middleware.cors import CORSMiddleware -from pydantic import BaseModel +from flask import Flask, request, send_from_directory, Response +from flask_cors import CORS +from flask_socketio import SocketIO, emit, disconnect from generator import load_csm_1b, Segment -import uvicorn import time import gc from collections import deque +from threading import Lock # Select device if torch.cuda.is_available(): @@ -36,73 +32,39 @@ print("Loading WhisperX model...") asr_model = whisperx.load_model("medium", device, compute_type="float16") print("WhisperX model loaded!") -app = FastAPI() - -# Add CORS middleware to allow cross-origin requests -app.add_middleware( - CORSMiddleware, - allow_origins=["*"], # Allow all origins in development - allow_credentials=True, - allow_methods=["*"], - allow_headers=["*"], -) +# Silence detection parameters +SILENCE_THRESHOLD = 0.01 # Adjust based on your audio normalization +SILENCE_DURATION_SEC = 1.0 # How long silence must persist # Define the base directory base_dir = os.path.dirname(os.path.abspath(__file__)) - -# Mount a static files directory if you have any static assets like CSS or JS static_dir = os.path.join(base_dir, "static") -os.makedirs(static_dir, exist_ok=True) # Create the directory if it doesn't exist -app.mount("/static", StaticFiles(directory=static_dir), name="static") +os.makedirs(static_dir, exist_ok=True) -# Define route to serve index.html as the main page -@app.get("/", response_class=HTMLResponse) -async def get_index(): - try: - with open(os.path.join(base_dir, "index.html"), "r") as f: - return HTMLResponse(content=f.read()) - except FileNotFoundError: - return HTMLResponse(content="

Error: index.html not found

") +# Setup Flask +app = Flask(__name__) +CORS(app) +socketio = SocketIO(app, cors_allowed_origins="*", async_mode='eventlet') -# Add a favicon endpoint (optional, but good to have) -@app.get("/favicon.ico") -async def get_favicon(): - favicon_path = os.path.join(static_dir, "favicon.ico") - if os.path.exists(favicon_path): - return FileResponse(favicon_path) - else: - return HTMLResponse(status_code=204) # No content - -# Connection manager to handle multiple clients -class ConnectionManager: - def __init__(self): - self.active_connections: List[WebSocket] = [] - - async def connect(self, websocket: WebSocket): - await websocket.accept() - self.active_connections.append(websocket) - - def disconnect(self, websocket: WebSocket): - self.active_connections.remove(websocket) - -manager = ConnectionManager() - -# Silence detection parameters -SILENCE_THRESHOLD = 0.01 # Adjust based on your audio normalization -SILENCE_DURATION_SEC = 1.0 # How long silence must persist to be considered "stopped talking" +# Socket connection management +thread = None +thread_lock = Lock() +active_clients = {} # Map client_id to client context # Helper function to convert audio data -async def decode_audio_data(audio_data: str) -> torch.Tensor: +def decode_audio_data(audio_data: str) -> torch.Tensor: """Decode base64 audio data to a torch tensor""" try: + # Extract the actual base64 content + if ',' in audio_data: + audio_data = audio_data.split(',')[1] + # Decode base64 audio data - binary_data = base64.b64decode(audio_data.split(',')[1] if ',' in audio_data else audio_data) + binary_data = base64.b64decode(audio_data) - # Save to a temporary WAV file first - temp_file = BytesIO(binary_data) - - # Load audio from binary data, explicitly specifying the format - audio_tensor, sample_rate = torchaudio.load(temp_file, format="wav") + # Load audio from binary data + with BytesIO(binary_data) as temp_file: + audio_tensor, sample_rate = torchaudio.load(temp_file, format="wav") # Resample if needed if sample_rate != generator.sample_rate: @@ -121,7 +83,7 @@ async def decode_audio_data(audio_data: str) -> torch.Tensor: return torch.zeros(generator.sample_rate // 2) # 0.5 seconds of silence -async def encode_audio_data(audio_tensor: torch.Tensor) -> str: +def encode_audio_data(audio_tensor: torch.Tensor) -> str: """Encode torch tensor audio to base64 string""" buf = BytesIO() torchaudio.save(buf, audio_tensor.unsqueeze(0).cpu(), generator.sample_rate, format="wav") @@ -130,40 +92,36 @@ async def encode_audio_data(audio_tensor: torch.Tensor) -> str: return f"data:audio/wav;base64,{audio_base64}" -async def transcribe_audio(audio_tensor: torch.Tensor) -> str: +def transcribe_audio(audio_tensor: torch.Tensor) -> str: """Transcribe audio using WhisperX""" try: # Save the tensor to a temporary file - temp_file = BytesIO() - torchaudio.save(temp_file, audio_tensor.unsqueeze(0).cpu(), generator.sample_rate, format="wav") - temp_file.seek(0) - - # Create a temporary file on disk (WhisperX requires a file path) - temp_path = "temp_audio.wav" - with open(temp_path, "wb") as f: - f.write(temp_file.read()) + temp_path = os.path.join(base_dir, "temp_audio.wav") + torchaudio.save(temp_path, audio_tensor.unsqueeze(0).cpu(), generator.sample_rate) # Load and transcribe the audio audio = whisperx.load_audio(temp_path) result = asr_model.transcribe(audio, batch_size=16) # Clean up - os.remove(temp_path) + if os.path.exists(temp_path): + os.remove(temp_path) # Get the transcription text if result["segments"] and len(result["segments"]) > 0: # Combine all segments transcription = " ".join([segment["text"] for segment in result["segments"]]) - print(f"Transcription: {transcription}") return transcription.strip() else: return "" except Exception as e: print(f"Error in transcription: {str(e)}") + if os.path.exists("temp_audio.wav"): + os.remove("temp_audio.wav") return "" -async def generate_response(text: str, conversation_history: List[Segment]) -> str: +def generate_response(text: str, conversation_history: List[Segment]) -> str: """Generate a contextual response based on the transcribed text""" # Simple response logic - can be replaced with a more sophisticated LLM in the future responses = { @@ -191,311 +149,319 @@ async def generate_response(text: str, conversation_history: List[Segment]) -> s else: return f"I understand you said '{text}'. That's interesting! Can you tell me more about that?" +# Flask routes for serving static content +@app.route('/') +def index(): + return send_from_directory(base_dir, 'index.html') -@app.websocket("/ws") -async def websocket_endpoint(websocket: WebSocket): - await manager.connect(websocket) - context_segments = [] # Store conversation context - streaming_buffer = [] # Buffer for streaming audio chunks - is_streaming = False +@app.route('/favicon.ico') +def favicon(): + if os.path.exists(os.path.join(static_dir, 'favicon.ico')): + return send_from_directory(static_dir, 'favicon.ico') + return Response(status=204) + +@app.route('/static/') +def serve_static(path): + return send_from_directory(static_dir, path) + +# Socket.IO event handlers +@socketio.on('connect') +def handle_connect(): + client_id = request.sid + print(f"Client connected: {client_id}") - # Variables for silence detection - last_active_time = time.time() - is_silence = False - energy_window = deque(maxlen=10) # For tracking recent audio energy + # Initialize client context + active_clients[client_id] = { + 'context_segments': [], + 'streaming_buffer': [], + 'is_streaming': False, + 'is_silence': False, + 'last_active_time': time.time(), + 'energy_window': deque(maxlen=10) + } + + emit('status', {'type': 'connected', 'message': 'Connected to server'}) + +@socketio.on('disconnect') +def handle_disconnect(): + client_id = request.sid + if client_id in active_clients: + del active_clients[client_id] + print(f"Client disconnected: {client_id}") + +@socketio.on('generate') +def handle_generate(data): + client_id = request.sid + if client_id not in active_clients: + emit('error', {'message': 'Client not registered'}) + return try: - while True: - # Receive JSON data from client - data = await websocket.receive_text() - request = json.loads(data) - - action = request.get("action") - - if action == "generate": - try: - text = request.get("text", "") - speaker_id = request.get("speaker", 0) - - # Generate audio response - print(f"Generating audio for: '{text}' with speaker {speaker_id}") - audio_tensor = generator.generate( - text=text, - speaker=speaker_id, - context=context_segments, - max_audio_length_ms=10_000, - ) - - # Add to conversation context - context_segments.append(Segment(text=text, speaker=speaker_id, audio=audio_tensor)) - - # Convert audio to base64 and send back to client - audio_base64 = await encode_audio_data(audio_tensor) - await websocket.send_json({ - "type": "audio_response", - "audio": audio_base64 - }) - except Exception as e: - print(f"Error generating audio: {str(e)}") - await websocket.send_json({ - "type": "error", - "message": f"Error generating audio: {str(e)}" - }) - - elif action == "add_to_context": - try: - text = request.get("text", "") - speaker_id = request.get("speaker", 0) - audio_data = request.get("audio", "") - - # Convert received audio to tensor - audio_tensor = await decode_audio_data(audio_data) - - # Add to conversation context - context_segments.append(Segment(text=text, speaker=speaker_id, audio=audio_tensor)) - - await websocket.send_json({ - "type": "context_updated", - "message": "Audio added to context" - }) - except Exception as e: - print(f"Error adding to context: {str(e)}") - await websocket.send_json({ - "type": "error", - "message": f"Error processing audio: {str(e)}" - }) - - elif action == "clear_context": - context_segments = [] - await websocket.send_json({ - "type": "context_updated", - "message": "Context cleared" - }) - - elif action == "stream_audio": - try: - speaker_id = request.get("speaker", 0) - audio_data = request.get("audio", "") - - # Convert received audio to tensor - audio_chunk = await decode_audio_data(audio_data) - - # Start streaming mode if not already started - if not is_streaming: - is_streaming = True - streaming_buffer = [] - energy_window.clear() - is_silence = False - last_active_time = time.time() - print(f"Streaming started with speaker ID: {speaker_id}") - await websocket.send_json({ - "type": "streaming_status", - "status": "started" - }) - - # Calculate audio energy for silence detection - chunk_energy = torch.mean(torch.abs(audio_chunk)).item() - energy_window.append(chunk_energy) - avg_energy = sum(energy_window) / len(energy_window) - - # Debug audio levels - if len(energy_window) >= 5: # Only start printing after we have enough samples - if avg_energy > SILENCE_THRESHOLD: - print(f"[AUDIO] Active sound detected - Energy: {avg_energy:.6f} (threshold: {SILENCE_THRESHOLD})") - else: - print(f"[AUDIO] Silence detected - Energy: {avg_energy:.6f} (threshold: {SILENCE_THRESHOLD})") - - # Check if audio is silent - current_silence = avg_energy < SILENCE_THRESHOLD - - # Track silence transition - if not is_silence and current_silence: - # Transition to silence - is_silence = True - last_active_time = time.time() - print("[STREAM] Transition to silence detected") - elif is_silence and not current_silence: - # User started talking again - is_silence = False - print("[STREAM] User resumed speaking") - - # Add chunk to buffer regardless of silence state - streaming_buffer.append(audio_chunk) - - # Debug buffer size periodically - if len(streaming_buffer) % 10 == 0: - print(f"[BUFFER] Current size: {len(streaming_buffer)} chunks, ~{len(streaming_buffer)/5:.1f} seconds") - - # Check if silence has persisted long enough to consider "stopped talking" - silence_elapsed = time.time() - last_active_time - - if is_silence and silence_elapsed >= SILENCE_DURATION_SEC and len(streaming_buffer) > 0: - # User has stopped talking - process the collected audio - print(f"[STREAM] Processing audio after {silence_elapsed:.2f}s of silence") - print(f"[STREAM] Processing {len(streaming_buffer)} audio chunks (~{len(streaming_buffer)/5:.1f} seconds)") - - full_audio = torch.cat(streaming_buffer, dim=0) - - # Log audio statistics - audio_duration = len(full_audio) / generator.sample_rate - audio_min = torch.min(full_audio).item() - audio_max = torch.max(full_audio).item() - audio_mean = torch.mean(full_audio).item() - print(f"[AUDIO] Processed audio - Duration: {audio_duration:.2f}s, Min: {audio_min:.4f}, Max: {audio_max:.4f}, Mean: {audio_mean:.4f}") - - # Process with WhisperX speech-to-text - print("[ASR] Starting transcription with WhisperX...") - transcribed_text = await transcribe_audio(full_audio) - - # Log the transcription - print(f"[ASR] Transcribed text: '{transcribed_text}'") - - # Add to conversation context - if transcribed_text: - print(f"[DIALOG] Adding user utterance to context: '{transcribed_text}'") - user_segment = Segment(text=transcribed_text, speaker=speaker_id, audio=full_audio) - context_segments.append(user_segment) - - # Generate a contextual response - print("[DIALOG] Generating response...") - response_text = await generate_response(transcribed_text, context_segments) - print(f"[DIALOG] Response text: '{response_text}'") - - # Send the transcribed text to client - await websocket.send_json({ - "type": "transcription", - "text": transcribed_text - }) - - # Generate audio for the response - print("[TTS] Generating speech for response...") - audio_tensor = generator.generate( - text=response_text, - speaker=1 if speaker_id == 0 else 0, # Use opposite speaker - context=context_segments, - max_audio_length_ms=10_000, - ) - print(f"[TTS] Generated audio length: {len(audio_tensor)/generator.sample_rate:.2f}s") - - # Add response to context - ai_segment = Segment( - text=response_text, - speaker=1 if speaker_id == 0 else 0, - audio=audio_tensor - ) - context_segments.append(ai_segment) - print(f"[DIALOG] Context now has {len(context_segments)} segments") - - # Convert audio to base64 and send back to client - audio_base64 = await encode_audio_data(audio_tensor) - print("[STREAM] Sending audio response to client") - await websocket.send_json({ - "type": "audio_response", - "text": response_text, - "audio": audio_base64 - }) - else: - print("[ASR] Transcription failed or returned empty text") - # If transcription failed, send a generic response - await websocket.send_json({ - "type": "error", - "message": "Sorry, I couldn't understand what you said. Could you try again?" - }) - - # Clear buffer and reset silence detection - streaming_buffer = [] - energy_window.clear() - is_silence = False - last_active_time = time.time() - print("[STREAM] Buffer cleared, ready for next utterance") - - # If buffer gets too large without silence, process it anyway - # This prevents memory issues with very long streams - elif len(streaming_buffer) >= 30: # ~6 seconds of audio at 5 chunks/sec - print("[BUFFER] Maximum buffer size reached, processing audio") - full_audio = torch.cat(streaming_buffer, dim=0) - - # Process with WhisperX speech-to-text - print("[ASR] Starting forced transcription of long audio...") - transcribed_text = await transcribe_audio(full_audio) - - if transcribed_text: - print(f"[ASR] Transcribed long audio: '{transcribed_text}'") - context_segments.append(Segment(text=transcribed_text, speaker=speaker_id, audio=full_audio)) - - # Send the transcribed text to client - await websocket.send_json({ - "type": "transcription", - "text": transcribed_text + " (processing continued speech...)" - }) - else: - print("[ASR] No transcription from long audio") - - streaming_buffer = [] - print("[BUFFER] Buffer cleared due to size limit") - - except Exception as e: - print(f"[ERROR] Processing streaming audio: {str(e)}") - # Print traceback for more detailed error information - import traceback - traceback.print_exc() - await websocket.send_json({ - "type": "error", - "message": f"Error processing streaming audio: {str(e)}" - }) - - elif action == "stop_streaming": - is_streaming = False - if streaming_buffer and len(streaming_buffer) > 5: # Only process if there's meaningful audio - # Process any remaining audio in the buffer - full_audio = torch.cat(streaming_buffer, dim=0) - - # Process with WhisperX speech-to-text - transcribed_text = await transcribe_audio(full_audio) - - if transcribed_text: - context_segments.append(Segment(text=transcribed_text, speaker=request.get("speaker", 0), audio=full_audio)) - - # Send the transcribed text to client - await websocket.send_json({ - "type": "transcription", - "text": transcribed_text - }) - - streaming_buffer = [] - await websocket.send_json({ - "type": "streaming_status", - "status": "stopped" - }) - - except WebSocketDisconnect: - manager.disconnect(websocket) - print("Client disconnected") + text = data.get('text', '') + speaker_id = data.get('speaker', 0) + + print(f"Generating audio for: '{text}' with speaker {speaker_id}") + + # Generate audio response + audio_tensor = generator.generate( + text=text, + speaker=speaker_id, + context=active_clients[client_id]['context_segments'], + max_audio_length_ms=10_000, + ) + + # Add to conversation context + active_clients[client_id]['context_segments'].append( + Segment(text=text, speaker=speaker_id, audio=audio_tensor) + ) + + # Convert audio to base64 and send back to client + audio_base64 = encode_audio_data(audio_tensor) + emit('audio_response', { + 'type': 'audio_response', + 'audio': audio_base64 + }) + except Exception as e: - print(f"Error: {str(e)}") - try: - await websocket.send_json({ - "type": "error", - "message": str(e) - }) - except: - pass - manager.disconnect(websocket) + print(f"Error generating audio: {str(e)}") + emit('error', { + 'type': 'error', + 'message': f"Error generating audio: {str(e)}" + }) + +@socketio.on('add_to_context') +def handle_add_to_context(data): + client_id = request.sid + if client_id not in active_clients: + emit('error', {'message': 'Client not registered'}) + return + + try: + text = data.get('text', '') + speaker_id = data.get('speaker', 0) + audio_data = data.get('audio', '') + + # Convert received audio to tensor + audio_tensor = decode_audio_data(audio_data) + + # Add to conversation context + active_clients[client_id]['context_segments'].append( + Segment(text=text, speaker=speaker_id, audio=audio_tensor) + ) + + emit('context_updated', { + 'type': 'context_updated', + 'message': 'Audio added to context' + }) + + except Exception as e: + print(f"Error adding to context: {str(e)}") + emit('error', { + 'type': 'error', + 'message': f"Error processing audio: {str(e)}" + }) + +@socketio.on('clear_context') +def handle_clear_context(): + client_id = request.sid + if client_id in active_clients: + active_clients[client_id]['context_segments'] = [] + + emit('context_updated', { + 'type': 'context_updated', + 'message': 'Context cleared' + }) + +@socketio.on('stream_audio') +def handle_stream_audio(data): + client_id = request.sid + if client_id not in active_clients: + emit('error', {'message': 'Client not registered'}) + return + + client = active_clients[client_id] + + try: + speaker_id = data.get('speaker', 0) + audio_data = data.get('audio', '') + + # Convert received audio to tensor + audio_chunk = decode_audio_data(audio_data) + + # Start streaming mode if not already started + if not client['is_streaming']: + client['is_streaming'] = True + client['streaming_buffer'] = [] + client['energy_window'].clear() + client['is_silence'] = False + client['last_active_time'] = time.time() + print(f"[{client_id}] Streaming started with speaker ID: {speaker_id}") + emit('streaming_status', { + 'type': 'streaming_status', + 'status': 'started' + }) + + # Calculate audio energy for silence detection + chunk_energy = torch.mean(torch.abs(audio_chunk)).item() + client['energy_window'].append(chunk_energy) + avg_energy = sum(client['energy_window']) / len(client['energy_window']) + + # Check if audio is silent + current_silence = avg_energy < SILENCE_THRESHOLD + + # Track silence transition + if not client['is_silence'] and current_silence: + # Transition to silence + client['is_silence'] = True + client['last_active_time'] = time.time() + elif client['is_silence'] and not current_silence: + # User started talking again + client['is_silence'] = False + + # Add chunk to buffer regardless of silence state + client['streaming_buffer'].append(audio_chunk) + + # Check if silence has persisted long enough to consider "stopped talking" + silence_elapsed = time.time() - client['last_active_time'] + + if client['is_silence'] and silence_elapsed >= SILENCE_DURATION_SEC and len(client['streaming_buffer']) > 0: + # User has stopped talking - process the collected audio + print(f"[{client_id}] Processing audio after {silence_elapsed:.2f}s of silence") + + full_audio = torch.cat(client['streaming_buffer'], dim=0) + + # Process with WhisperX speech-to-text + print(f"[{client_id}] Starting transcription with WhisperX...") + transcribed_text = transcribe_audio(full_audio) + + # Log the transcription + print(f"[{client_id}] Transcribed text: '{transcribed_text}'") + + # Add to conversation context + if transcribed_text: + user_segment = Segment(text=transcribed_text, speaker=speaker_id, audio=full_audio) + client['context_segments'].append(user_segment) + + # Generate a contextual response + response_text = generate_response(transcribed_text, client['context_segments']) + + # Send the transcribed text to client + emit('transcription', { + 'type': 'transcription', + 'text': transcribed_text + }) + + # Generate audio for the response + audio_tensor = generator.generate( + text=response_text, + speaker=1 if speaker_id == 0 else 0, # Use opposite speaker + context=client['context_segments'], + max_audio_length_ms=10_000, + ) + + # Add response to context + ai_segment = Segment( + text=response_text, + speaker=1 if speaker_id == 0 else 0, + audio=audio_tensor + ) + client['context_segments'].append(ai_segment) + + # Convert audio to base64 and send back to client + audio_base64 = encode_audio_data(audio_tensor) + emit('audio_response', { + 'type': 'audio_response', + 'text': response_text, + 'audio': audio_base64 + }) + else: + # If transcription failed, send a generic response + emit('error', { + 'type': 'error', + 'message': "Sorry, I couldn't understand what you said. Could you try again?" + }) + + # Clear buffer and reset silence detection + client['streaming_buffer'] = [] + client['energy_window'].clear() + client['is_silence'] = False + client['last_active_time'] = time.time() + + # If buffer gets too large without silence, process it anyway + elif len(client['streaming_buffer']) >= 30: # ~6 seconds of audio at 5 chunks/sec + full_audio = torch.cat(client['streaming_buffer'], dim=0) + + # Process with WhisperX speech-to-text + transcribed_text = transcribe_audio(full_audio) + + if transcribed_text: + client['context_segments'].append( + Segment(text=transcribed_text, speaker=speaker_id, audio=full_audio) + ) + + # Send the transcribed text to client + emit('transcription', { + 'type': 'transcription', + 'text': transcribed_text + " (processing continued speech...)" + }) + + client['streaming_buffer'] = [] + + except Exception as e: + import traceback + traceback.print_exc() + print(f"Error processing streaming audio: {str(e)}") + emit('error', { + 'type': 'error', + 'message': f"Error processing streaming audio: {str(e)}" + }) + +@socketio.on('stop_streaming') +def handle_stop_streaming(data): + client_id = request.sid + if client_id not in active_clients: + return + + client = active_clients[client_id] + client['is_streaming'] = False + + if client['streaming_buffer'] and len(client['streaming_buffer']) > 5: + # Process any remaining audio in the buffer + full_audio = torch.cat(client['streaming_buffer'], dim=0) + + # Process with WhisperX speech-to-text + transcribed_text = transcribe_audio(full_audio) + + if transcribed_text: + client['context_segments'].append( + Segment(text=transcribed_text, speaker=data.get("speaker", 0), audio=full_audio) + ) + + # Send the transcribed text to client + emit('transcription', { + 'type': 'transcription', + 'text': transcribed_text + }) + + client['streaming_buffer'] = [] + emit('streaming_status', { + 'type': 'streaming_status', + 'status': 'stopped' + }) -# Update the __main__ block with a comprehensive server startup message if __name__ == "__main__": print(f"\n{'='*60}") - print(f"🔊 Sesame AI Voice Chat Server") + print(f"🔊 Sesame AI Voice Chat Server (Flask Implementation)") print(f"{'='*60}") print(f"📡 Server Information:") - print(f" - Local URL: http://localhost:8000") - print(f" - Network URL: http://:8000") - print(f" - WebSocket: ws://:8000/ws") + print(f" - Local URL: http://localhost:5000") + print(f" - Network URL: http://:5000") + print(f" - WebSocket: ws://:5000/socket.io") print(f"{'='*60}") print(f"💡 To make this server public:") - print(f" 1. Ensure port 8000 is open in your firewall") - print(f" 2. Set up port forwarding on your router to port 8000") - print(f" 3. Or use a service like ngrok with: ngrok http 8000") + print(f" 1. Ensure port 5000 is open in your firewall") + print(f" 2. Set up port forwarding on your router to port 5000") + print(f" 3. Or use a service like ngrok with: ngrok http 5000") print(f"{'='*60}") print(f"🌐 Device: {device.upper()}") print(f"🧠 Models loaded: Sesame CSM + WhisperX ({asr_model.device})") @@ -503,5 +469,4 @@ if __name__ == "__main__": print(f"{'='*60}") print(f"Ready to receive connections! Press Ctrl+C to stop the server.\n") - # Start the server - uvicorn.run(app, host="0.0.0.0", port=8000) \ No newline at end of file + socketio.run(app, host="0.0.0.0", port=5000, debug=False) \ No newline at end of file From 14c08bc93edffe68bbe25f88e9078d26043e26c3 Mon Sep 17 00:00:00 2001 From: GamerBoss101 Date: Sat, 29 Mar 2025 23:14:20 -0400 Subject: [PATCH 04/10] Demo Frontend Update --- Backend/index.html | 1157 +++++++++++++++-------------------------- Backend/voice-chat.js | 795 ++++++++++++++++++++++++++++ 2 files changed, 1219 insertions(+), 733 deletions(-) create mode 100644 Backend/voice-chat.js diff --git a/Backend/index.html b/Backend/index.html index 2944700..cbb4172 100644 --- a/Backend/index.html +++ b/Backend/index.html @@ -1,801 +1,492 @@ -/Backend/index.html --> Sesame AI Voice Chat - + -

Sesame AI Voice Chat

-
- -
- -
Audio levels will appear here when speaking
-
- -
- - - -
- -
-
-
Not connected
+
+

Sesame AI Voice Chat

+

Speak naturally and have a conversation with AI

+
+ +
+
+
+

Conversation

+ +
+
+
+ +
+
+

Audio Visualizer

+
+ +
Speak to see audio visualization
+
+
+ +
+
+
Voice Settings
+ + +
+
+ Silence Threshold + 0.01 +
+ +
+ +
+
+
+
+ +
+
Conversation Controls
+
+ +
+
+
+ +
+
Settings
+
+
+ + + Auto-play responses +
+
+ + + Show visualizer +
+
+
+ +
+
+
Not connected
+
+
- +
+

Powered by Sesame AI | WhisperX for speech recognition

+
+ + + \ No newline at end of file diff --git a/Backend/voice-chat.js b/Backend/voice-chat.js new file mode 100644 index 0000000..0c8a815 --- /dev/null +++ b/Backend/voice-chat.js @@ -0,0 +1,795 @@ +/** + * Sesame AI Voice Chat Application + * + * This script handles the audio streaming, visualization, + * and Socket.IO communication for the voice chat application. + */ + +// Application state +const state = { + socket: null, + audioContext: null, + streamProcessor: null, + analyser: null, + microphone: null, + isStreaming: false, + isSpeaking: false, + silenceTimer: null, + energyWindow: [], + currentSpeaker: 0, + silenceThreshold: 0.01, + visualizerAnimationFrame: null, + volumeUpdateInterval: null, + connectionAttempts: 0 +}; + +// Constants +const ENERGY_WINDOW_SIZE = 10; +const CLIENT_SILENCE_DURATION_MS = 1000; // 1 second of silence before processing +const MAX_CONNECTION_ATTEMPTS = 5; +const RECONNECTION_DELAY_MS = 2000; + +// DOM elements +const elements = { + conversation: document.getElementById('conversation'), + speakerSelect: document.getElementById('speakerSelect'), + streamButton: document.getElementById('streamButton'), + clearButton: document.getElementById('clearButton'), + statusDot: document.getElementById('statusDot'), + statusText: document.getElementById('statusText'), + visualizerCanvas: document.getElementById('audioVisualizer'), + visualizerLabel: document.getElementById('visualizerLabel'), + thresholdSlider: document.getElementById('thresholdSlider'), + thresholdValue: document.getElementById('thresholdValue'), + volumeLevel: document.getElementById('volumeLevel'), + autoPlayResponses: document.getElementById('autoPlayResponses'), + showVisualizer: document.getElementById('showVisualizer') +}; + +// Visualization variables +let canvasContext; +let visualizerBufferLength; +let visualizerDataArray; + +// Initialize the application +function initializeApp() { + // Set up event listeners + elements.streamButton.addEventListener('click', toggleStreaming); + elements.clearButton.addEventListener('click', clearConversation); + elements.thresholdSlider.addEventListener('input', updateThreshold); + elements.speakerSelect.addEventListener('change', () => { + state.currentSpeaker = parseInt(elements.speakerSelect.value); + }); + elements.showVisualizer.addEventListener('change', toggleVisualizerVisibility); + + // Initialize audio context + setupAudioContext(); + + // Set up visualization + setupVisualizer(); + + // Connect to Socket.IO server + connectToServer(); + + // Add welcome message + addSystemMessage('Welcome to Sesame AI Voice Chat! Click "Start Conversation" to begin speaking.'); +} + +// Connect to Socket.IO server +function connectToServer() { + try { + // Use the server URL with or without a specific port + const serverUrl = window.location.origin; + + updateStatus('Connecting...', 'connecting'); + console.log(`Connecting to Socket.IO server at ${serverUrl}`); + + state.socket = io(serverUrl, { + reconnectionDelay: RECONNECTION_DELAY_MS, + reconnectionDelayMax: 5000, + reconnectionAttempts: MAX_CONNECTION_ATTEMPTS + }); + + setupSocketListeners(); + } catch (error) { + console.error('Error connecting to server:', error); + updateStatus('Connection failed. Retrying...', 'error'); + + // Try to reconnect + if (state.connectionAttempts < MAX_CONNECTION_ATTEMPTS) { + state.connectionAttempts++; + setTimeout(connectToServer, RECONNECTION_DELAY_MS); + } else { + updateStatus('Could not connect to server', 'error'); + addSystemMessage('Failed to connect to the server. Please check your connection and refresh the page.'); + } + } +} + +// Set up Socket.IO event listeners +function setupSocketListeners() { + if (!state.socket) return; + + state.socket.on('connect', () => { + console.log('Connected to Socket.IO server'); + updateStatus('Connected', 'connected'); + state.connectionAttempts = 0; + elements.streamButton.disabled = false; + addSystemMessage('Connected to server'); + }); + + state.socket.on('disconnect', () => { + console.log('Disconnected from Socket.IO server'); + updateStatus('Disconnected', 'disconnected'); + + // Stop streaming if active + if (state.isStreaming) { + stopStreaming(false); // false = don't send to server + } + + elements.streamButton.disabled = true; + addSystemMessage('Disconnected from server. Trying to reconnect...'); + }); + + state.socket.on('status', (data) => { + console.log('Status:', data); + addSystemMessage(data.message); + }); + + state.socket.on('error', (data) => { + console.error('Server error:', data); + addSystemMessage(`Error: ${data.message}`); + }); + + state.socket.on('audio_response', handleAudioResponse); + state.socket.on('transcription', handleTranscription); + state.socket.on('context_updated', handleContextUpdate); + state.socket.on('streaming_status', handleStreamingStatus); + + state.socket.on('connect_error', (error) => { + console.error('Connection error:', error); + updateStatus('Connection Error', 'error'); + }); +} + +// Update the connection status in the UI +function updateStatus(message, status) { + elements.statusText.textContent = message; + elements.statusDot.className = 'status-dot'; + + if (status === 'connected') { + elements.statusDot.classList.add('active'); + } else if (status === 'connecting') { + elements.statusDot.style.backgroundColor = '#FFA500'; + } else if (status === 'error') { + elements.statusDot.style.backgroundColor = '#F44336'; + } +} + +// Set up audio context +function setupAudioContext() { + try { + state.audioContext = new (window.AudioContext || window.webkitAudioContext)(); + console.log('Audio context initialized'); + } catch (err) { + console.error('Error setting up audio context:', err); + addSystemMessage(`Audio context error: ${err.message}`); + elements.streamButton.disabled = true; + } +} + +// Set up audio visualizer +function setupVisualizer() { + canvasContext = elements.visualizerCanvas.getContext('2d'); + + // Set canvas size to match container + function resizeCanvas() { + const container = elements.visualizerCanvas.parentElement; + elements.visualizerCanvas.width = container.clientWidth; + elements.visualizerCanvas.height = container.clientHeight; + } + + // Call initially and on window resize + resizeCanvas(); + window.addEventListener('resize', resizeCanvas); + + // Create placeholder data array + visualizerBufferLength = 128; + visualizerDataArray = new Uint8Array(visualizerBufferLength); +} + +// Toggle stream on/off +function toggleStreaming() { + if (state.isStreaming) { + stopStreaming(true); // true = send to server + } else { + startStreaming(); + } +} + +// Start streaming audio to the server +async function startStreaming() { + if (!state.socket || !state.socket.connected) { + addSystemMessage('Cannot start conversation: Not connected to server'); + return; + } + + try { + // Request microphone access + const stream = await navigator.mediaDevices.getUserMedia({ audio: true }); + + // Update state + state.isStreaming = true; + state.isSpeaking = false; + state.energyWindow = []; + state.currentSpeaker = parseInt(elements.speakerSelect.value); + + // Update UI + elements.streamButton.innerHTML = ' Listening...'; + elements.streamButton.classList.add('recording'); + elements.visualizerLabel.style.opacity = '0'; + + // Set up audio processing + setupAudioProcessing(stream); + + // Start volume meter updates + state.volumeUpdateInterval = setInterval(updateVolumeMeter, 100); + + addSystemMessage('Listening - speak naturally and pause when finished'); + + } catch (err) { + console.error('Error starting audio stream:', err); + addSystemMessage(`Microphone error: ${err.message}`); + cleanupAudioResources(); + } +} + +// Set up audio processing pipeline +function setupAudioProcessing(stream) { + // Store microphone stream for later cleanup + state.microphone = stream; + + // Create source from microphone + const source = state.audioContext.createMediaStreamSource(stream); + + // Setup analyzer for visualization + state.analyser = state.audioContext.createAnalyser(); + state.analyser.fftSize = 256; + state.analyser.smoothingTimeConstant = 0.8; + state.analyser.minDecibels = -90; + state.analyser.maxDecibels = -10; + + visualizerBufferLength = state.analyser.frequencyBinCount; + visualizerDataArray = new Uint8Array(visualizerBufferLength); + + // Connect source to analyzer + source.connect(state.analyser); + + // Start visualization + if (state.visualizerAnimationFrame) { + cancelAnimationFrame(state.visualizerAnimationFrame); + } + drawVisualizer(); + + // Setup audio processor + state.streamProcessor = state.audioContext.createScriptProcessor(4096, 1, 1); + + // Connect audio nodes + source.connect(state.streamProcessor); + state.streamProcessor.connect(state.audioContext.destination); + + // Process audio + state.streamProcessor.onaudioprocess = handleAudioProcess; +} + +// Handle each frame of audio data +function handleAudioProcess(e) { + const audioData = e.inputBuffer.getChannelData(0); + + // Calculate energy (volume) for silence detection + const energy = calculateAudioEnergy(audioData); + updateEnergyWindow(energy); + + // Check if currently silent + const avgEnergy = calculateAverageEnergy(); + const isSilent = avgEnergy < state.silenceThreshold; + + // Handle silence/speech transitions + handleSpeechState(isSilent); + + // Process and send audio + const downsampled = downsampleBuffer(audioData, state.audioContext.sampleRate, 24000); + sendAudioChunk(downsampled, state.currentSpeaker); +} + +// Stop streaming audio +function stopStreaming(sendToServer = true) { + // Cleanup audio resources + cleanupAudioResources(); + + // Reset state + state.isStreaming = false; + state.isSpeaking = false; + state.energyWindow = []; + + // Update UI + elements.streamButton.innerHTML = ' Start Conversation'; + elements.streamButton.classList.remove('recording', 'processing'); + elements.streamButton.style.backgroundColor = ''; + elements.volumeLevel.style.width = '100%'; + + // Clear volume meter updates + if (state.volumeUpdateInterval) { + clearInterval(state.volumeUpdateInterval); + state.volumeUpdateInterval = null; + } + + addSystemMessage('Conversation paused'); + + // Notify server + if (sendToServer && state.socket && state.socket.connected) { + state.socket.emit('stop_streaming', { + speaker: state.currentSpeaker + }); + } +} + +// Clean up audio processing resources +function cleanupAudioResources() { + // Stop microphone stream + if (state.microphone) { + state.microphone.getTracks().forEach(track => track.stop()); + state.microphone = null; + } + + // Disconnect audio processor + if (state.streamProcessor) { + state.streamProcessor.disconnect(); + state.streamProcessor.onaudioprocess = null; + state.streamProcessor = null; + } + + // Disconnect analyzer + if (state.analyser) { + state.analyser.disconnect(); + state.analyser = null; + } + + // Cancel visualizer animation + if (state.visualizerAnimationFrame) { + cancelAnimationFrame(state.visualizerAnimationFrame); + state.visualizerAnimationFrame = null; + } + + // Cancel silence timer + if (state.silenceTimer) { + clearTimeout(state.silenceTimer); + state.silenceTimer = null; + } + + // Reset visualizer display + if (canvasContext) { + canvasContext.clearRect(0, 0, elements.visualizerCanvas.width, elements.visualizerCanvas.height); + elements.visualizerLabel.style.opacity = '0.7'; + } +} + +// Clear conversation history +function clearConversation() { + // Clear UI + elements.conversation.innerHTML = ''; + addSystemMessage('Conversation cleared'); + + // Notify server + if (state.socket && state.socket.connected) { + state.socket.emit('clear_context'); + } +} + +// Calculate audio energy (volume) +function calculateAudioEnergy(buffer) { + let sum = 0; + for (let i = 0; i < buffer.length; i++) { + sum += Math.abs(buffer[i]); + } + return sum / buffer.length; +} + +// Update energy window for averaging +function updateEnergyWindow(energy) { + state.energyWindow.push(energy); + if (state.energyWindow.length > ENERGY_WINDOW_SIZE) { + state.energyWindow.shift(); + } +} + +// Calculate average energy from window +function calculateAverageEnergy() { + if (state.energyWindow.length === 0) return 0; + return state.energyWindow.reduce((sum, val) => sum + val, 0) / state.energyWindow.length; +} + +// Update the threshold from the slider +function updateThreshold() { + state.silenceThreshold = parseFloat(elements.thresholdSlider.value); + elements.thresholdValue.textContent = state.silenceThreshold.toFixed(3); +} + +// Update the volume meter display +function updateVolumeMeter() { + if (!state.isStreaming || !state.analyser) return; + + // Get current volume level + const dataArray = new Uint8Array(state.analyser.frequencyBinCount); + state.analyser.getByteFrequencyData(dataArray); + + // Calculate average volume + let sum = 0; + for (let i = 0; i < dataArray.length; i++) { + sum += dataArray[i]; + } + const average = sum / dataArray.length; + + // Normalize to 0-100% + const percentage = Math.min(100, Math.max(0, average / 128 * 100)); + + // Invert because we're showing the "empty" portion + elements.volumeLevel.style.width = (100 - percentage) + '%'; + + // Change color based on level + if (percentage > 70) { + elements.volumeLevel.style.backgroundColor = 'rgba(244, 67, 54, 0.5)'; // Red + } else if (percentage > 30) { + elements.volumeLevel.style.backgroundColor = 'rgba(255, 235, 59, 0.5)'; // Yellow + } else { + elements.volumeLevel.style.backgroundColor = 'rgba(0, 0, 0, 0.5)'; // Dark + } +} + +// Handle speech/silence state transitions +function handleSpeechState(isSilent) { + if (state.isSpeaking && isSilent) { + // Transition from speaking to silence + if (!state.silenceTimer) { + state.silenceTimer = setTimeout(() => { + // Silence persisted long enough - process the audio + elements.streamButton.innerHTML = ' Processing...'; + elements.streamButton.classList.remove('recording'); + elements.streamButton.classList.add('processing'); + addSystemMessage('Detected pause in speech, processing response...'); + }, CLIENT_SILENCE_DURATION_MS); + } + } else if (!state.isSpeaking && !isSilent) { + // Transition from silence to speaking + state.isSpeaking = true; + elements.streamButton.innerHTML = ' Listening...'; + elements.streamButton.classList.add('recording'); + elements.streamButton.classList.remove('processing'); + + // Clear silence timer + if (state.silenceTimer) { + clearTimeout(state.silenceTimer); + state.silenceTimer = null; + } + } else if (state.isSpeaking && !isSilent) { + // Still speaking, reset silence timer + if (state.silenceTimer) { + clearTimeout(state.silenceTimer); + state.silenceTimer = null; + } + } + + // Update speaking state for non-silent audio + if (!isSilent) { + state.isSpeaking = true; + } +} + +// Send audio chunk to server +function sendAudioChunk(audioData, speaker) { + if (!state.socket || !state.socket.connected) { + console.warn('Cannot send audio: socket not connected'); + return; + } + + const wavData = createWavBlob(audioData, 24000); + const reader = new FileReader(); + + reader.onloadend = function() { + const base64data = reader.result; + + // Send to server using Socket.IO + state.socket.emit('stream_audio', { + speaker: speaker, + audio: base64data + }); + }; + + reader.readAsDataURL(wavData); +} + +// Draw audio visualizer +function drawVisualizer() { + if (!canvasContext) { + return; + } + + state.visualizerAnimationFrame = requestAnimationFrame(drawVisualizer); + + // Skip drawing if visualizer is hidden + if (!elements.showVisualizer.checked) { + if (elements.visualizerCanvas.style.opacity !== '0') { + elements.visualizerCanvas.style.opacity = '0'; + } + return; + } else if (elements.visualizerCanvas.style.opacity !== '1') { + elements.visualizerCanvas.style.opacity = '1'; + } + + // Get frequency data if available + if (state.isStreaming && state.analyser) { + try { + state.analyser.getByteFrequencyData(visualizerDataArray); + } catch (e) { + console.error("Error getting frequency data:", e); + } + } else { + // Fade out when not streaming + for (let i = 0; i < visualizerDataArray.length; i++) { + visualizerDataArray[i] = Math.max(0, visualizerDataArray[i] - 5); + } + } + + // Clear canvas + canvasContext.fillStyle = 'rgb(0, 0, 0)'; + canvasContext.fillRect(0, 0, elements.visualizerCanvas.width, elements.visualizerCanvas.height); + + // Draw gradient bars + const width = elements.visualizerCanvas.width; + const height = elements.visualizerCanvas.height; + const barCount = Math.min(visualizerBufferLength, 64); + const barWidth = width / barCount - 1; + + for (let i = 0; i < barCount; i++) { + const index = Math.floor(i * visualizerBufferLength / barCount); + const value = visualizerDataArray[index]; + + // Use logarithmic scale for better audio visualization + // This makes low values more visible while still maintaining full range + const logFactor = 20; + const scaledValue = Math.log(1 + (value / 255) * logFactor) / Math.log(1 + logFactor); + const barHeight = scaledValue * height; + + // Position bars + const x = i * (barWidth + 1); + const y = height - barHeight; + + // Create color gradient based on frequency and amplitude + const hue = i / barCount * 360; // Full color spectrum + const saturation = 80 + (value / 255 * 20); // Higher values more saturated + const lightness = 40 + (value / 255 * 20); // Dynamic brightness based on amplitude + + // Draw main bar + canvasContext.fillStyle = `hsl(${hue}, ${saturation}%, ${lightness}%)`; + canvasContext.fillRect(x, y, barWidth, barHeight); + + // Add reflection effect + if (barHeight > 5) { + const gradient = canvasContext.createLinearGradient( + x, y, + x, y + barHeight * 0.5 + ); + gradient.addColorStop(0, `hsla(${hue}, ${saturation}%, ${lightness + 20}%, 0.4)`); + gradient.addColorStop(1, `hsla(${hue}, ${saturation}%, ${lightness}%, 0)`); + canvasContext.fillStyle = gradient; + canvasContext.fillRect(x, y, barWidth, barHeight * 0.5); + + // Add highlight on top of the bar for better 3D effect + canvasContext.fillStyle = `hsla(${hue}, ${saturation - 20}%, ${lightness + 30}%, 0.7)`; + canvasContext.fillRect(x, y, barWidth, 2); + } + } + + // Show/hide the label + elements.visualizerLabel.style.opacity = (state.isStreaming) ? '0' : '0.7'; +} + +// Toggle visualizer visibility +function toggleVisualizerVisibility() { + const isVisible = elements.showVisualizer.checked; + elements.visualizerCanvas.style.opacity = isVisible ? '1' : '0'; + + if (isVisible && state.isStreaming && !state.visualizerAnimationFrame) { + drawVisualizer(); + } +} + +// Handle audio response from server +function handleAudioResponse(data) { + console.log('Received audio response'); + + // Create message container + const messageElement = document.createElement('div'); + messageElement.className = 'message ai'; + + // Add text content if available + if (data.text) { + const textElement = document.createElement('p'); + textElement.textContent = data.text; + messageElement.appendChild(textElement); + } + + // Create and configure audio element + const audioElement = document.createElement('audio'); + audioElement.controls = true; + audioElement.className = 'audio-player'; + + // Set audio source + const audioSource = document.createElement('source'); + audioSource.src = data.audio; + audioSource.type = 'audio/wav'; + + // Add fallback text + audioElement.textContent = 'Your browser does not support the audio element.'; + + // Assemble audio element + audioElement.appendChild(audioSource); + messageElement.appendChild(audioElement); + + // Add timestamp + const timeElement = document.createElement('span'); + timeElement.className = 'message-time'; + timeElement.textContent = new Date().toLocaleTimeString(); + messageElement.appendChild(timeElement); + + // Add to conversation + elements.conversation.appendChild(messageElement); + + // Auto-scroll to bottom + elements.conversation.scrollTop = elements.conversation.scrollHeight; + + // Auto-play if enabled + if (elements.autoPlayResponses.checked) { + audioElement.play() + .catch(err => { + console.warn('Auto-play failed:', err); + addSystemMessage('Auto-play failed. Please click play to hear the response.'); + }); + } + + // Re-enable stream button after processing is complete + if (state.isStreaming) { + elements.streamButton.innerHTML = ' Listening...'; + elements.streamButton.classList.add('recording'); + elements.streamButton.classList.remove('processing'); + } +} + +// Handle transcription response from server +function handleTranscription(data) { + console.log('Received transcription:', data.text); + + // Create message element + const messageElement = document.createElement('div'); + messageElement.className = 'message user'; + + // Add text content + const textElement = document.createElement('p'); + textElement.textContent = data.text; + messageElement.appendChild(textElement); + + // Add timestamp + const timeElement = document.createElement('span'); + timeElement.className = 'message-time'; + timeElement.textContent = new Date().toLocaleTimeString(); + messageElement.appendChild(timeElement); + + // Add to conversation + elements.conversation.appendChild(messageElement); + + // Auto-scroll to bottom + elements.conversation.scrollTop = elements.conversation.scrollHeight; +} + +// Handle context update from server +function handleContextUpdate(data) { + console.log('Context updated:', data.message); +} + +// Handle streaming status updates from server +function handleStreamingStatus(data) { + console.log('Streaming status:', data.status); + + if (data.status === 'stopped') { + // Reset UI if needed + if (state.isStreaming) { + stopStreaming(false); // Don't send to server since this came from server + } + } +} + +// Add a system message to the conversation +function addSystemMessage(message) { + const messageElement = document.createElement('div'); + messageElement.className = 'message system'; + messageElement.textContent = message; + elements.conversation.appendChild(messageElement); + + // Auto-scroll to bottom + elements.conversation.scrollTop = elements.conversation.scrollHeight; +} + +// Create WAV blob from audio data +function createWavBlob(audioData, sampleRate) { + // Function to convert Float32Array to Int16Array for WAV format + function floatTo16BitPCM(output, offset, input) { + for (let i = 0; i < input.length; i++, offset += 2) { + const s = Math.max(-1, Math.min(1, input[i])); + output.setInt16(offset, s < 0 ? s * 0x8000 : s * 0x7FFF, true); + } + } + + // Create WAV header + function writeString(view, offset, string) { + for (let i = 0; i < string.length; i++) { + view.setUint8(offset + i, string.charCodeAt(i)); + } + } + + // Create WAV file with header + function encodeWAV(samples) { + const buffer = new ArrayBuffer(44 + samples.length * 2); + const view = new DataView(buffer); + + // RIFF chunk descriptor + writeString(view, 0, 'RIFF'); + view.setUint32(4, 36 + samples.length * 2, true); + writeString(view, 8, 'WAVE'); + + // fmt sub-chunk + writeString(view, 12, 'fmt '); + view.setUint32(16, 16, true); + view.setUint16(20, 1, true); // PCM format + view.setUint16(22, 1, true); // Mono channel + view.setUint32(24, sampleRate, true); + view.setUint32(28, sampleRate * 2, true); // Byte rate + view.setUint16(32, 2, true); // Block align + view.setUint16(34, 16, true); // Bits per sample + + // data sub-chunk + writeString(view, 36, 'data'); + view.setUint32(40, samples.length * 2, true); + floatTo16BitPCM(view, 44, samples); + + return buffer; + } + + // Convert audio data to TypedArray if it's a regular Array + const samples = Array.isArray(audioData) ? new Float32Array(audioData) : audioData; + + // Create WAV blob + const wavBuffer = encodeWAV(samples); + return new Blob([wavBuffer], { type: 'audio/wav' }); +} + +// Downsample audio buffer to target sample rate +function downsampleBuffer(buffer, originalSampleRate, targetSampleRate) { + if (originalSampleRate === targetSampleRate) { + return buffer; + } + + const ratio = originalSampleRate / targetSampleRate; + const newLength = Math.round(buffer.length / ratio); + const result = new Float32Array(newLength); + + for (let i = 0; i < newLength; i++) { + const pos = Math.round(i * ratio); + result[i] = buffer[pos]; + } + + return result; +} + +// Initialize the application when DOM is fully loaded +document.addEventListener('DOMContentLoaded', initializeApp); + From 9ca259aab3e6c16060f8a2343db5bd76b50230ad Mon Sep 17 00:00:00 2001 From: GamerBoss101 Date: Sat, 29 Mar 2025 23:22:45 -0400 Subject: [PATCH 05/10] Demo Update 2 --- Backend/index.html | 2 +- Backend/voice-chat.js | 708 ++++++++++++++++++++++-------------------- 2 files changed, 374 insertions(+), 336 deletions(-) diff --git a/Backend/index.html b/Backend/index.html index cbb4172..5ea925c 100644 --- a/Backend/index.html +++ b/Backend/index.html @@ -487,6 +487,6 @@ - + \ No newline at end of file diff --git a/Backend/voice-chat.js b/Backend/voice-chat.js index 0c8a815..a4e10f5 100644 --- a/Backend/voice-chat.js +++ b/Backend/voice-chat.js @@ -1,388 +1,445 @@ /** - * Sesame AI Voice Chat Application + * Sesame AI Voice Chat Client * - * This script handles the audio streaming, visualization, - * and Socket.IO communication for the voice chat application. + * A web client that connects to a Sesame AI voice chat server and enables + * real-time voice conversation with an AI assistant. */ +// Configuration constants +const SERVER_URL = window.location.hostname === 'localhost' ? + 'http://localhost:5000' : window.location.origin; +const ENERGY_WINDOW_SIZE = 15; +const CLIENT_SILENCE_DURATION_MS = 750; + +// DOM elements +const elements = { + conversation: null, + streamButton: null, + clearButton: null, + thresholdSlider: null, + thresholdValue: null, + visualizerCanvas: null, + visualizerLabel: null, + volumeLevel: null, + statusDot: null, + statusText: null, + speakerSelection: null, + autoPlayResponses: null, + showVisualizer: null +}; + // Application state const state = { socket: null, audioContext: null, - streamProcessor: null, analyser: null, microphone: null, + streamProcessor: null, isStreaming: false, isSpeaking: false, - silenceTimer: null, - energyWindow: [], - currentSpeaker: 0, silenceThreshold: 0.01, - visualizerAnimationFrame: null, + energyWindow: [], + silenceTimer: null, volumeUpdateInterval: null, - connectionAttempts: 0 + visualizerAnimationFrame: null, + currentSpeaker: 0 }; -// Constants -const ENERGY_WINDOW_SIZE = 10; -const CLIENT_SILENCE_DURATION_MS = 1000; // 1 second of silence before processing -const MAX_CONNECTION_ATTEMPTS = 5; -const RECONNECTION_DELAY_MS = 2000; - -// DOM elements -const elements = { - conversation: document.getElementById('conversation'), - speakerSelect: document.getElementById('speakerSelect'), - streamButton: document.getElementById('streamButton'), - clearButton: document.getElementById('clearButton'), - statusDot: document.getElementById('statusDot'), - statusText: document.getElementById('statusText'), - visualizerCanvas: document.getElementById('audioVisualizer'), - visualizerLabel: document.getElementById('visualizerLabel'), - thresholdSlider: document.getElementById('thresholdSlider'), - thresholdValue: document.getElementById('thresholdValue'), - volumeLevel: document.getElementById('volumeLevel'), - autoPlayResponses: document.getElementById('autoPlayResponses'), - showVisualizer: document.getElementById('showVisualizer') -}; - -// Visualization variables -let canvasContext; -let visualizerBufferLength; -let visualizerDataArray; +// Visualizer variables +let canvasContext = null; +let visualizerBufferLength = 0; +let visualizerDataArray = null; // Initialize the application function initializeApp() { - // Set up event listeners - elements.streamButton.addEventListener('click', toggleStreaming); - elements.clearButton.addEventListener('click', clearConversation); - elements.thresholdSlider.addEventListener('input', updateThreshold); - elements.speakerSelect.addEventListener('change', () => { - state.currentSpeaker = parseInt(elements.speakerSelect.value); - }); - elements.showVisualizer.addEventListener('change', toggleVisualizerVisibility); - - // Initialize audio context - setupAudioContext(); + // Initialize the UI elements + initializeUIElements(); - // Set up visualization + // Initialize socket.io connection + setupSocketConnection(); + + // Setup event listeners + setupEventListeners(); + + // Initialize visualizer setupVisualizer(); - // Connect to Socket.IO server - connectToServer(); - - // Add welcome message - addSystemMessage('Welcome to Sesame AI Voice Chat! Click "Start Conversation" to begin speaking.'); + // Show welcome message + addSystemMessage('Welcome to Sesame AI Voice Chat! Click "Start Conversation" to begin.'); } -// Connect to Socket.IO server -function connectToServer() { - try { - // Use the server URL with or without a specific port - const serverUrl = window.location.origin; +// Initialize UI elements +function initializeUIElements() { + // Main UI containers + const chatContainer = document.querySelector('.chat-container'); + const controlPanel = document.querySelector('.control-panel'); + + // Create conversation section + chatContainer.innerHTML = ` +
+

Conversation

+
+
+ Disconnected +
+
+
+ `; + + // Create control panel + controlPanel.innerHTML = ` +
+
+ +
Speak to see audio visualization
+
+
- updateStatus('Connecting...', 'connecting'); - console.log(`Connecting to Socket.IO server at ${serverUrl}`); - - state.socket = io(serverUrl, { - reconnectionDelay: RECONNECTION_DELAY_MS, - reconnectionDelayMax: 5000, - reconnectionAttempts: MAX_CONNECTION_ATTEMPTS - }); - - setupSocketListeners(); - } catch (error) { - console.error('Error connecting to server:', error); - updateStatus('Connection failed. Retrying...', 'error'); - - // Try to reconnect - if (state.connectionAttempts < MAX_CONNECTION_ATTEMPTS) { - state.connectionAttempts++; - setTimeout(connectToServer, RECONNECTION_DELAY_MS); - } else { - updateStatus('Could not connect to server', 'error'); - addSystemMessage('Failed to connect to the server. Please check your connection and refresh the page.'); - } - } +
+
+
Voice Controls
+ +
+
+
+ +
+
+ Silence Threshold + 0.01 +
+ +
+ + + +
+ + +
+
+ +
+
Settings
+ +
+
+ + +
+ +
+ + +
+
+
+
+ `; + + // Store references to UI elements + elements.conversation = document.querySelector('.conversation'); + elements.streamButton = document.getElementById('streamButton'); + elements.clearButton = document.getElementById('clearButton'); + elements.thresholdSlider = document.getElementById('thresholdSlider'); + elements.thresholdValue = document.getElementById('thresholdValue'); + elements.visualizerCanvas = document.getElementById('audioVisualizer'); + elements.visualizerLabel = document.querySelector('.visualizer-label'); + elements.volumeLevel = document.querySelector('.volume-level'); + elements.statusDot = document.querySelector('.status-dot'); + elements.statusText = document.querySelector('.status-text'); + elements.speakerSelection = document.getElementById('speakerSelection'); + elements.autoPlayResponses = document.getElementById('autoPlayResponses'); + elements.showVisualizer = document.getElementById('showVisualizer'); } -// Set up Socket.IO event listeners -function setupSocketListeners() { - if (!state.socket) return; +// Setup Socket.IO connection +function setupSocketConnection() { + state.socket = io(SERVER_URL); + // Connection events state.socket.on('connect', () => { - console.log('Connected to Socket.IO server'); - updateStatus('Connected', 'connected'); - state.connectionAttempts = 0; - elements.streamButton.disabled = false; - addSystemMessage('Connected to server'); + console.log('Connected to server'); + updateConnectionStatus(true); }); state.socket.on('disconnect', () => { - console.log('Disconnected from Socket.IO server'); - updateStatus('Disconnected', 'disconnected'); + console.log('Disconnected from server'); + updateConnectionStatus(false); // Stop streaming if active if (state.isStreaming) { - stopStreaming(false); // false = don't send to server + stopStreaming(false); } - - elements.streamButton.disabled = true; - addSystemMessage('Disconnected from server. Trying to reconnect...'); - }); - - state.socket.on('status', (data) => { - console.log('Status:', data); - addSystemMessage(data.message); }); state.socket.on('error', (data) => { - console.error('Server error:', data); + console.error('Socket error:', data.message); addSystemMessage(`Error: ${data.message}`); }); + // Register message handlers state.socket.on('audio_response', handleAudioResponse); state.socket.on('transcription', handleTranscription); state.socket.on('context_updated', handleContextUpdate); state.socket.on('streaming_status', handleStreamingStatus); +} + +// Setup event listeners +function setupEventListeners() { + // Stream button + elements.streamButton.addEventListener('click', toggleStreaming); - state.socket.on('connect_error', (error) => { - console.error('Connection error:', error); - updateStatus('Connection Error', 'error'); + // Clear button + elements.clearButton.addEventListener('click', clearConversation); + + // Threshold slider + elements.thresholdSlider.addEventListener('input', updateThreshold); + + // Speaker selection + elements.speakerSelection.addEventListener('change', () => { + state.currentSpeaker = parseInt(elements.speakerSelection.value, 10); }); -} - -// Update the connection status in the UI -function updateStatus(message, status) { - elements.statusText.textContent = message; - elements.statusDot.className = 'status-dot'; - if (status === 'connected') { - elements.statusDot.classList.add('active'); - } else if (status === 'connecting') { - elements.statusDot.style.backgroundColor = '#FFA500'; - } else if (status === 'error') { - elements.statusDot.style.backgroundColor = '#F44336'; - } + // Visualizer toggle + elements.showVisualizer.addEventListener('change', toggleVisualizerVisibility); } -// Set up audio context -function setupAudioContext() { - try { - state.audioContext = new (window.AudioContext || window.webkitAudioContext)(); - console.log('Audio context initialized'); - } catch (err) { - console.error('Error setting up audio context:', err); - addSystemMessage(`Audio context error: ${err.message}`); - elements.streamButton.disabled = true; - } -} - -// Set up audio visualizer +// Setup audio visualizer function setupVisualizer() { + if (!elements.visualizerCanvas) return; + canvasContext = elements.visualizerCanvas.getContext('2d'); - // Set canvas size to match container - function resizeCanvas() { - const container = elements.visualizerCanvas.parentElement; - elements.visualizerCanvas.width = container.clientWidth; - elements.visualizerCanvas.height = container.clientHeight; - } + // Set canvas dimensions + elements.visualizerCanvas.width = elements.visualizerCanvas.offsetWidth; + elements.visualizerCanvas.height = elements.visualizerCanvas.offsetHeight; - // Call initially and on window resize - resizeCanvas(); - window.addEventListener('resize', resizeCanvas); - - // Create placeholder data array - visualizerBufferLength = 128; - visualizerDataArray = new Uint8Array(visualizerBufferLength); + // Initialize the visualizer + drawVisualizer(); } -// Toggle stream on/off +// Update connection status UI +function updateConnectionStatus(isConnected) { + elements.statusDot.classList.toggle('active', isConnected); + elements.statusText.textContent = isConnected ? 'Connected' : 'Disconnected'; +} + +// Toggle streaming state function toggleStreaming() { if (state.isStreaming) { - stopStreaming(true); // true = send to server + stopStreaming(true); } else { startStreaming(); } } // Start streaming audio to the server -async function startStreaming() { - if (!state.socket || !state.socket.connected) { - addSystemMessage('Cannot start conversation: Not connected to server'); - return; - } +function startStreaming() { + if (state.isStreaming) return; - try { - // Request microphone access - const stream = await navigator.mediaDevices.getUserMedia({ audio: true }); - - // Update state - state.isStreaming = true; - state.isSpeaking = false; - state.energyWindow = []; - state.currentSpeaker = parseInt(elements.speakerSelect.value); - - // Update UI - elements.streamButton.innerHTML = ' Listening...'; - elements.streamButton.classList.add('recording'); - elements.visualizerLabel.style.opacity = '0'; - - // Set up audio processing - setupAudioProcessing(stream); - - // Start volume meter updates - state.volumeUpdateInterval = setInterval(updateVolumeMeter, 100); - - addSystemMessage('Listening - speak naturally and pause when finished'); - - } catch (err) { - console.error('Error starting audio stream:', err); - addSystemMessage(`Microphone error: ${err.message}`); - cleanupAudioResources(); - } -} - -// Set up audio processing pipeline -function setupAudioProcessing(stream) { - // Store microphone stream for later cleanup - state.microphone = stream; - - // Create source from microphone - const source = state.audioContext.createMediaStreamSource(stream); - - // Setup analyzer for visualization - state.analyser = state.audioContext.createAnalyser(); - state.analyser.fftSize = 256; - state.analyser.smoothingTimeConstant = 0.8; - state.analyser.minDecibels = -90; - state.analyser.maxDecibels = -10; - - visualizerBufferLength = state.analyser.frequencyBinCount; - visualizerDataArray = new Uint8Array(visualizerBufferLength); - - // Connect source to analyzer - source.connect(state.analyser); - - // Start visualization - if (state.visualizerAnimationFrame) { - cancelAnimationFrame(state.visualizerAnimationFrame); - } - drawVisualizer(); - - // Setup audio processor - state.streamProcessor = state.audioContext.createScriptProcessor(4096, 1, 1); - - // Connect audio nodes - source.connect(state.streamProcessor); - state.streamProcessor.connect(state.audioContext.destination); - - // Process audio - state.streamProcessor.onaudioprocess = handleAudioProcess; -} - -// Handle each frame of audio data -function handleAudioProcess(e) { - const audioData = e.inputBuffer.getChannelData(0); - - // Calculate energy (volume) for silence detection - const energy = calculateAudioEnergy(audioData); - updateEnergyWindow(energy); - - // Check if currently silent - const avgEnergy = calculateAverageEnergy(); - const isSilent = avgEnergy < state.silenceThreshold; - - // Handle silence/speech transitions - handleSpeechState(isSilent); - - // Process and send audio - const downsampled = downsampleBuffer(audioData, state.audioContext.sampleRate, 24000); - sendAudioChunk(downsampled, state.currentSpeaker); + // Request microphone access + navigator.mediaDevices.getUserMedia({ audio: true, video: false }) + .then(stream => { + // Show processing state while setting up + elements.streamButton.innerHTML = ' Initializing...'; + + // Create audio context + state.audioContext = new (window.AudioContext || window.webkitAudioContext)(); + + // Create microphone source + state.microphone = state.audioContext.createMediaStreamSource(stream); + + // Create analyser for visualizer + state.analyser = state.audioContext.createAnalyser(); + state.analyser.fftSize = 256; + visualizerBufferLength = state.analyser.frequencyBinCount; + visualizerDataArray = new Uint8Array(visualizerBufferLength); + + // Connect microphone to analyser + state.microphone.connect(state.analyser); + + // Create script processor for audio processing + const bufferSize = 4096; + state.streamProcessor = state.audioContext.createScriptProcessor(bufferSize, 1, 1); + + // Set up audio processing callback + state.streamProcessor.onaudioprocess = handleAudioProcess; + + // Connect the processors + state.analyser.connect(state.streamProcessor); + state.streamProcessor.connect(state.audioContext.destination); + + // Update UI + state.isStreaming = true; + elements.streamButton.innerHTML = ' Listening...'; + elements.streamButton.classList.add('recording'); + + // Initialize energy window + state.energyWindow = []; + + // Start volume meter updates + state.volumeUpdateInterval = setInterval(updateVolumeMeter, 100); + + // Start visualizer if enabled + if (elements.showVisualizer.checked && !state.visualizerAnimationFrame) { + drawVisualizer(); + } + + // Show starting message + addSystemMessage('Listening... Speak clearly into your microphone.'); + + // Notify the server that we're starting + state.socket.emit('stream_audio', { + audio: '', + speaker: state.currentSpeaker + }); + }) + .catch(err => { + console.error('Error accessing microphone:', err); + addSystemMessage(`Error: ${err.message}. Please make sure your microphone is connected and you've granted permission.`); + elements.streamButton.innerHTML = ' Start Conversation'; + }); } // Stop streaming audio -function stopStreaming(sendToServer = true) { - // Cleanup audio resources - cleanupAudioResources(); +function stopStreaming(notifyServer = true) { + if (!state.isStreaming) return; - // Reset state - state.isStreaming = false; - state.isSpeaking = false; - state.energyWindow = []; - - // Update UI + // Update UI first elements.streamButton.innerHTML = ' Start Conversation'; - elements.streamButton.classList.remove('recording', 'processing'); - elements.streamButton.style.backgroundColor = ''; - elements.volumeLevel.style.width = '100%'; + elements.streamButton.classList.remove('recording'); + elements.streamButton.classList.remove('processing'); - // Clear volume meter updates + // Stop volume meter updates if (state.volumeUpdateInterval) { clearInterval(state.volumeUpdateInterval); state.volumeUpdateInterval = null; } - addSystemMessage('Conversation paused'); - - // Notify server - if (sendToServer && state.socket && state.socket.connected) { - state.socket.emit('stop_streaming', { - speaker: state.currentSpeaker - }); - } -} - -// Clean up audio processing resources -function cleanupAudioResources() { - // Stop microphone stream - if (state.microphone) { - state.microphone.getTracks().forEach(track => track.stop()); - state.microphone = null; - } - - // Disconnect audio processor + // Stop all audio processing if (state.streamProcessor) { state.streamProcessor.disconnect(); - state.streamProcessor.onaudioprocess = null; state.streamProcessor = null; } - // Disconnect analyzer if (state.analyser) { state.analyser.disconnect(); - state.analyser = null; } - // Cancel visualizer animation + if (state.microphone) { + state.microphone.disconnect(); + } + + // Close audio context + if (state.audioContext && state.audioContext.state !== 'closed') { + state.audioContext.close().catch(err => console.warn('Error closing audio context:', err)); + } + + // Cleanup animation frames + if (state.visualizerAnimationFrame) { + cancelAnimationFrame(state.visualizerAnimationFrame); + state.visualizerAnimationFrame = null; + } + + // Reset state + state.isStreaming = false; + state.isSpeaking = false; + + // Notify the server + if (notifyServer && state.socket && state.socket.connected) { + state.socket.emit('stop_streaming', { + speaker: state.currentSpeaker + }); + } + + // Show message + addSystemMessage('Conversation paused. Click "Start Conversation" to resume.'); +} + +// Handle audio processing +function handleAudioProcess(event) { + const inputData = event.inputBuffer.getChannelData(0); + + // Calculate audio energy (volume level) + const energy = calculateAudioEnergy(inputData); + + // Update energy window for averaging + updateEnergyWindow(energy); + + // Calculate average energy + const avgEnergy = calculateAverageEnergy(); + + // Determine if audio is silent + const isSilent = avgEnergy < state.silenceThreshold; + + // Handle speech state based on silence + handleSpeechState(isSilent); + + // Only send audio chunk if we detect speech + if (!isSilent) { + // Create a resampled version at 24kHz for the server + // Most WebRTC audio is 48kHz, but we want 24kHz for the model + const resampledData = downsampleBuffer(inputData, state.audioContext.sampleRate, 24000); + + // Send the audio chunk to the server + sendAudioChunk(resampledData, state.currentSpeaker); + } +} + +// Cleanup audio resources when done +function cleanupAudioResources() { + // Stop all audio processing + if (state.streamProcessor) { + state.streamProcessor.disconnect(); + state.streamProcessor = null; + } + + if (state.analyser) { + state.analyser.disconnect(); + state.analyser = null; + } + + if (state.microphone) { + state.microphone.disconnect(); + state.microphone = null; + } + + // Close audio context + if (state.audioContext && state.audioContext.state !== 'closed') { + state.audioContext.close().catch(err => console.warn('Error closing audio context:', err)); + } + + // Cancel all timers and animation frames + if (state.volumeUpdateInterval) { + clearInterval(state.volumeUpdateInterval); + state.volumeUpdateInterval = null; + } + if (state.visualizerAnimationFrame) { cancelAnimationFrame(state.visualizerAnimationFrame); state.visualizerAnimationFrame = null; } - // Cancel silence timer if (state.silenceTimer) { clearTimeout(state.silenceTimer); state.silenceTimer = null; } - - // Reset visualizer display - if (canvasContext) { - canvasContext.clearRect(0, 0, elements.visualizerCanvas.width, elements.visualizerCanvas.height); - elements.visualizerLabel.style.opacity = '0.7'; - } } // Clear conversation history function clearConversation() { - // Clear UI - elements.conversation.innerHTML = ''; - addSystemMessage('Conversation cleared'); - - // Notify server - if (state.socket && state.socket.connected) { - state.socket.emit('clear_context'); + if (elements.conversation) { + elements.conversation.innerHTML = ''; + addSystemMessage('Conversation cleared.'); + + // Notify server to clear context + if (state.socket && state.socket.connected) { + state.socket.emit('clear_context'); + } } } @@ -390,9 +447,9 @@ function clearConversation() { function calculateAudioEnergy(buffer) { let sum = 0; for (let i = 0; i < buffer.length; i++) { - sum += Math.abs(buffer[i]); + sum += buffer[i] * buffer[i]; } - return sum / buffer.length; + return Math.sqrt(sum / buffer.length); } // Update energy window for averaging @@ -406,7 +463,9 @@ function updateEnergyWindow(energy) { // Calculate average energy from window function calculateAverageEnergy() { if (state.energyWindow.length === 0) return 0; - return state.energyWindow.reduce((sum, val) => sum + val, 0) / state.energyWindow.length; + + const sum = state.energyWindow.reduce((a, b) => a + b, 0); + return sum / state.energyWindow.length; } // Update the threshold from the slider @@ -417,32 +476,26 @@ function updateThreshold() { // Update the volume meter display function updateVolumeMeter() { - if (!state.isStreaming || !state.analyser) return; + if (!state.isStreaming || !state.energyWindow.length) return; - // Get current volume level - const dataArray = new Uint8Array(state.analyser.frequencyBinCount); - state.analyser.getByteFrequencyData(dataArray); + const avgEnergy = calculateAverageEnergy(); - // Calculate average volume - let sum = 0; - for (let i = 0; i < dataArray.length; i++) { - sum += dataArray[i]; - } - const average = sum / dataArray.length; + // Scale energy to percentage (0-100) + // Typically, energy values will be very small (e.g., 0.001 to 0.1) + // So we multiply by a factor to make it more visible + const scaleFactor = 1000; + const percentage = Math.min(100, Math.max(0, avgEnergy * scaleFactor)); - // Normalize to 0-100% - const percentage = Math.min(100, Math.max(0, average / 128 * 100)); - - // Invert because we're showing the "empty" portion - elements.volumeLevel.style.width = (100 - percentage) + '%'; + // Update volume meter width + elements.volumeLevel.style.width = `${percentage}%`; // Change color based on level if (percentage > 70) { - elements.volumeLevel.style.backgroundColor = 'rgba(244, 67, 54, 0.5)'; // Red + elements.volumeLevel.style.backgroundColor = '#ff5252'; } else if (percentage > 30) { - elements.volumeLevel.style.backgroundColor = 'rgba(255, 235, 59, 0.5)'; // Yellow + elements.volumeLevel.style.backgroundColor = '#4CAF50'; } else { - elements.volumeLevel.style.backgroundColor = 'rgba(0, 0, 0, 0.5)'; // Dark + elements.volumeLevel.style.backgroundColor = '#4c84ff'; } } @@ -452,31 +505,16 @@ function handleSpeechState(isSilent) { // Transition from speaking to silence if (!state.silenceTimer) { state.silenceTimer = setTimeout(() => { - // Silence persisted long enough - process the audio - elements.streamButton.innerHTML = ' Processing...'; - elements.streamButton.classList.remove('recording'); - elements.streamButton.classList.add('processing'); - addSystemMessage('Detected pause in speech, processing response...'); + // Only consider it a real silence after a certain duration + // This prevents detecting brief pauses as the end of speech + state.isSpeaking = false; + state.silenceTimer = null; }, CLIENT_SILENCE_DURATION_MS); } - } else if (!state.isSpeaking && !isSilent) { - // Transition from silence to speaking - state.isSpeaking = true; - elements.streamButton.innerHTML = ' Listening...'; - elements.streamButton.classList.add('recording'); - elements.streamButton.classList.remove('processing'); - - // Clear silence timer - if (state.silenceTimer) { - clearTimeout(state.silenceTimer); - state.silenceTimer = null; - } - } else if (state.isSpeaking && !isSilent) { - // Still speaking, reset silence timer - if (state.silenceTimer) { - clearTimeout(state.silenceTimer); - state.silenceTimer = null; - } + } else if (state.silenceTimer && !isSilent) { + // User started speaking again, cancel the silence timer + clearTimeout(state.silenceTimer); + state.silenceTimer = null; } // Update speaking state for non-silent audio @@ -488,7 +526,7 @@ function handleSpeechState(isSilent) { // Send audio chunk to server function sendAudioChunk(audioData, speaker) { if (!state.socket || !state.socket.connected) { - console.warn('Cannot send audio: socket not connected'); + console.warn('Socket not connected'); return; } @@ -498,10 +536,10 @@ function sendAudioChunk(audioData, speaker) { reader.onloadend = function() { const base64data = reader.result; - // Send to server using Socket.IO + // Send the audio chunk to the server state.socket.emit('stream_audio', { - speaker: speaker, - audio: base64data + audio: base64data, + speaker: speaker }); }; @@ -531,7 +569,7 @@ function drawVisualizer() { try { state.analyser.getByteFrequencyData(visualizerDataArray); } catch (e) { - console.error("Error getting frequency data:", e); + console.warn('Error getting frequency data:', e); } } else { // Fade out when not streaming From 6a8cc50dac2eb99eb8095e13e261f846bfd7612f Mon Sep 17 00:00:00 2001 From: GamerBoss101 Date: Sat, 29 Mar 2025 23:28:44 -0400 Subject: [PATCH 06/10] serve voice chat js --- Backend/server.py | 4 ++++ 1 file changed, 4 insertions(+) diff --git a/Backend/server.py b/Backend/server.py index e986606..4e60aa7 100644 --- a/Backend/server.py +++ b/Backend/server.py @@ -160,6 +160,10 @@ def favicon(): return send_from_directory(static_dir, 'favicon.ico') return Response(status=204) +@app.route('/voice-chat.js') +def voice_chat_js(): + return send_from_directory(base_dir, 'voice-chat.js') + @app.route('/static/') def serve_static(path): return send_from_directory(static_dir, path) From b74ae2dbfc449913e669e2c54e76e973ad63eb6f Mon Sep 17 00:00:00 2001 From: GamerBoss101 Date: Sat, 29 Mar 2025 23:43:16 -0400 Subject: [PATCH 07/10] Demo Update 3 --- Backend/server.py | 62 ++++++++-- Backend/voice-chat.js | 275 +++++++++++++++++++++--------------------- 2 files changed, 188 insertions(+), 149 deletions(-) diff --git a/Backend/server.py b/Backend/server.py index 4e60aa7..bacf793 100644 --- a/Backend/server.py +++ b/Backend/server.py @@ -55,27 +55,71 @@ active_clients = {} # Map client_id to client context def decode_audio_data(audio_data: str) -> torch.Tensor: """Decode base64 audio data to a torch tensor""" try: + # Skip empty audio data + if not audio_data: + print("Empty audio data received") + return torch.zeros(generator.sample_rate // 2) # 0.5 seconds of silence + # Extract the actual base64 content if ',' in audio_data: audio_data = audio_data.split(',')[1] - + # Decode base64 audio data - binary_data = base64.b64decode(audio_data) + try: + binary_data = base64.b64decode(audio_data) + print(f"Decoded base64 data: {len(binary_data)} bytes") + except Exception as e: + print(f"Base64 decoding error: {str(e)}") + return torch.zeros(generator.sample_rate // 2) + # Debug: save the raw binary data to examine with external tools + debug_path = os.path.join(base_dir, "debug_incoming.wav") + with open(debug_path, 'wb') as f: + f.write(binary_data) + print(f"Saved debug file to {debug_path}") + # Load audio from binary data - with BytesIO(binary_data) as temp_file: - audio_tensor, sample_rate = torchaudio.load(temp_file, format="wav") + try: + with BytesIO(binary_data) as temp_file: + audio_tensor, sample_rate = torchaudio.load(temp_file, format="wav") + print(f"Loaded audio: shape={audio_tensor.shape}, sample_rate={sample_rate}Hz") + + # Check if audio is valid + if audio_tensor.numel() == 0 or torch.isnan(audio_tensor).any(): + print("Warning: Empty or invalid audio data detected") + return torch.zeros(generator.sample_rate // 2) + except Exception as e: + print(f"Audio loading error: {str(e)}") + # Try saving to a temporary file instead of loading from BytesIO + try: + temp_path = os.path.join(base_dir, "temp_incoming.wav") + with open(temp_path, 'wb') as f: + f.write(binary_data) + print(f"Trying to load from file: {temp_path}") + audio_tensor, sample_rate = torchaudio.load(temp_path, format="wav") + print(f"Loaded from file: shape={audio_tensor.shape}, sample_rate={sample_rate}Hz") + os.remove(temp_path) + except Exception as e2: + print(f"Secondary audio loading error: {str(e2)}") + return torch.zeros(generator.sample_rate // 2) # Resample if needed if sample_rate != generator.sample_rate: - audio_tensor = torchaudio.functional.resample( - audio_tensor.squeeze(0), - orig_freq=sample_rate, - new_freq=generator.sample_rate - ) + try: + print(f"Resampling from {sample_rate}Hz to {generator.sample_rate}Hz") + audio_tensor = torchaudio.functional.resample( + audio_tensor.squeeze(0), + orig_freq=sample_rate, + new_freq=generator.sample_rate + ) + print(f"Resampled audio shape: {audio_tensor.shape}") + except Exception as e: + print(f"Resampling error: {str(e)}") + return torch.zeros(generator.sample_rate // 2) else: audio_tensor = audio_tensor.squeeze(0) + print(f"Final audio tensor shape: {audio_tensor.shape}") return audio_tensor except Exception as e: print(f"Error decoding audio: {str(e)}") diff --git a/Backend/voice-chat.js b/Backend/voice-chat.js index a4e10f5..c85da8a 100644 --- a/Backend/voice-chat.js +++ b/Backend/voice-chat.js @@ -70,88 +70,18 @@ function initializeApp() { // Initialize UI elements function initializeUIElements() { - // Main UI containers - const chatContainer = document.querySelector('.chat-container'); - const controlPanel = document.querySelector('.control-panel'); - - // Create conversation section - chatContainer.innerHTML = ` -
-

Conversation

-
-
- Disconnected -
-
-
- `; - - // Create control panel - controlPanel.innerHTML = ` -
-
- -
Speak to see audio visualization
-
-
- -
-
-
Voice Controls
- -
-
-
- -
-
- Silence Threshold - 0.01 -
- -
- - - -
- - -
-
- -
-
Settings
- -
-
- - -
- -
- - -
-
-
-
- `; - // Store references to UI elements - elements.conversation = document.querySelector('.conversation'); + elements.conversation = document.getElementById('conversation'); elements.streamButton = document.getElementById('streamButton'); elements.clearButton = document.getElementById('clearButton'); elements.thresholdSlider = document.getElementById('thresholdSlider'); elements.thresholdValue = document.getElementById('thresholdValue'); elements.visualizerCanvas = document.getElementById('audioVisualizer'); - elements.visualizerLabel = document.querySelector('.visualizer-label'); - elements.volumeLevel = document.querySelector('.volume-level'); - elements.statusDot = document.querySelector('.status-dot'); - elements.statusText = document.querySelector('.status-text'); - elements.speakerSelection = document.getElementById('speakerSelection'); + elements.visualizerLabel = document.getElementById('visualizerLabel'); + elements.volumeLevel = document.getElementById('volumeLevel'); + elements.statusDot = document.getElementById('statusDot'); + elements.statusText = document.getElementById('statusText'); + elements.speakerSelection = document.getElementById('speakerSelect'); // Changed to match HTML elements.autoPlayResponses = document.getElementById('autoPlayResponses'); elements.showVisualizer = document.getElementById('showVisualizer'); } @@ -364,8 +294,12 @@ function stopStreaming(notifyServer = true) { function handleAudioProcess(event) { const inputData = event.inputBuffer.getChannelData(0); + // Log audio buffer statistics + console.log(`Audio buffer: length=${inputData.length}, sample rate=${state.audioContext.sampleRate}Hz`); + // Calculate audio energy (volume level) const energy = calculateAudioEnergy(inputData); + console.log(`Energy: ${energy.toFixed(6)}, threshold: ${state.silenceThreshold}`); // Update energy window for averaging updateEnergyWindow(energy); @@ -375,6 +309,7 @@ function handleAudioProcess(event) { // Determine if audio is silent const isSilent = avgEnergy < state.silenceThreshold; + console.log(`Silent: ${isSilent ? 'Yes' : 'No'}, avg energy: ${avgEnergy.toFixed(6)}`); // Handle speech state based on silence handleSpeechState(isSilent); @@ -384,6 +319,7 @@ function handleAudioProcess(event) { // Create a resampled version at 24kHz for the server // Most WebRTC audio is 48kHz, but we want 24kHz for the model const resampledData = downsampleBuffer(inputData, state.audioContext.sampleRate, 24000); + console.log(`Resampled audio: ${state.audioContext.sampleRate}Hz → 24000Hz, new length: ${resampledData.length}`); // Send the audio chunk to the server sendAudioChunk(resampledData, state.currentSpeaker); @@ -530,20 +466,132 @@ function sendAudioChunk(audioData, speaker) { return; } - const wavData = createWavBlob(audioData, 24000); - const reader = new FileReader(); + console.log(`Creating WAV from audio data: length=${audioData.length}`); - reader.onloadend = function() { - const base64data = reader.result; + // Check for NaN or invalid values + let hasNaN = false; + let min = Infinity; + let max = -Infinity; + let sum = 0; + + for (let i = 0; i < audioData.length; i++) { + if (isNaN(audioData[i]) || !isFinite(audioData[i])) { + hasNaN = true; + console.warn(`Invalid audio value at index ${i}: ${audioData[i]}`); + break; + } + min = Math.min(min, audioData[i]); + max = Math.max(max, audioData[i]); + sum += audioData[i]; + } + + if (hasNaN) { + console.warn('Audio data contains NaN or Infinity values. Creating silent audio instead.'); + audioData = new Float32Array(audioData.length).fill(0); + } else { + const avg = sum / audioData.length; + console.log(`Audio stats: min=${min.toFixed(4)}, max=${max.toFixed(4)}, avg=${avg.toFixed(4)}`); + } + + try { + // Create WAV blob with proper format + const wavData = createWavBlob(audioData, 24000); + console.log(`WAV blob created: size=${wavData.size} bytes, type=${wavData.type}`); - // Send the audio chunk to the server - state.socket.emit('stream_audio', { - audio: base64data, - speaker: speaker - }); - }; + const reader = new FileReader(); + + reader.onloadend = function() { + try { + // Get base64 data + const base64data = reader.result; + console.log(`Base64 data created: length=${base64data.length}`); + + // Validate the base64 data before sending + if (!base64data || base64data.length < 100) { + console.warn('Generated base64 data is too small or invalid'); + return; + } + + // Send the audio chunk to the server + console.log('Sending audio data to server...'); + state.socket.emit('stream_audio', { + audio: base64data, + speaker: speaker + }); + console.log('Audio data sent successfully'); + } catch (err) { + console.error('Error preparing audio data:', err); + } + }; + + reader.onerror = function(err) { + console.error('Error reading audio data:', err); + }; + + reader.readAsDataURL(wavData); + } catch (err) { + console.error('Error creating WAV data:', err); + } +} + +// Create WAV blob from audio data with validation +function createWavBlob(audioData, sampleRate) { + // Check if audio data is valid + if (!audioData || audioData.length === 0) { + console.warn('Empty audio data received'); + // Return a tiny silent audio snippet instead + audioData = new Float32Array(100).fill(0); + } - reader.readAsDataURL(wavData); + // Function to convert Float32Array to Int16Array for WAV format + function floatTo16BitPCM(output, offset, input) { + for (let i = 0; i < input.length; i++, offset += 2) { + const s = Math.max(-1, Math.min(1, input[i])); + output.setInt16(offset, s < 0 ? s * 0x8000 : s * 0x7FFF, true); + } + } + + // Create WAV header + function writeString(view, offset, string) { + for (let i = 0; i < string.length; i++) { + view.setUint8(offset + i, string.charCodeAt(i)); + } + } + + // Create WAV file with header + function encodeWAV(samples) { + const buffer = new ArrayBuffer(44 + samples.length * 2); + const view = new DataView(buffer); + + // RIFF chunk descriptor + writeString(view, 0, 'RIFF'); + view.setUint32(4, 36 + samples.length * 2, true); + writeString(view, 8, 'WAVE'); + + // fmt sub-chunk + writeString(view, 12, 'fmt '); + view.setUint32(16, 16, true); + view.setUint16(20, 1, true); // PCM format + view.setUint16(22, 1, true); // Mono channel + view.setUint32(24, sampleRate, true); + view.setUint32(28, sampleRate * 2, true); // Byte rate + view.setUint16(32, 2, true); // Block align + view.setUint16(34, 16, true); // Bits per sample + + // data sub-chunk + writeString(view, 36, 'data'); + view.setUint32(40, samples.length * 2, true); + floatTo16BitPCM(view, 44, samples); + + return buffer; + } + + // Convert audio data to TypedArray if it's a regular Array + const samples = Array.isArray(audioData) ? new Float32Array(audioData) : audioData; + + // Create WAV blob + const wavBuffer = encodeWAV(samples); + return new Blob([wavBuffer], { type: 'audio/wav' }); } // Draw audio visualizer @@ -757,59 +805,6 @@ function addSystemMessage(message) { elements.conversation.scrollTop = elements.conversation.scrollHeight; } -// Create WAV blob from audio data -function createWavBlob(audioData, sampleRate) { - // Function to convert Float32Array to Int16Array for WAV format - function floatTo16BitPCM(output, offset, input) { - for (let i = 0; i < input.length; i++, offset += 2) { - const s = Math.max(-1, Math.min(1, input[i])); - output.setInt16(offset, s < 0 ? s * 0x8000 : s * 0x7FFF, true); - } - } - - // Create WAV header - function writeString(view, offset, string) { - for (let i = 0; i < string.length; i++) { - view.setUint8(offset + i, string.charCodeAt(i)); - } - } - - // Create WAV file with header - function encodeWAV(samples) { - const buffer = new ArrayBuffer(44 + samples.length * 2); - const view = new DataView(buffer); - - // RIFF chunk descriptor - writeString(view, 0, 'RIFF'); - view.setUint32(4, 36 + samples.length * 2, true); - writeString(view, 8, 'WAVE'); - - // fmt sub-chunk - writeString(view, 12, 'fmt '); - view.setUint32(16, 16, true); - view.setUint16(20, 1, true); // PCM format - view.setUint16(22, 1, true); // Mono channel - view.setUint32(24, sampleRate, true); - view.setUint32(28, sampleRate * 2, true); // Byte rate - view.setUint16(32, 2, true); // Block align - view.setUint16(34, 16, true); // Bits per sample - - // data sub-chunk - writeString(view, 36, 'data'); - view.setUint32(40, samples.length * 2, true); - floatTo16BitPCM(view, 44, samples); - - return buffer; - } - - // Convert audio data to TypedArray if it's a regular Array - const samples = Array.isArray(audioData) ? new Float32Array(audioData) : audioData; - - // Create WAV blob - const wavBuffer = encodeWAV(samples); - return new Blob([wavBuffer], { type: 'audio/wav' }); -} - // Downsample audio buffer to target sample rate function downsampleBuffer(buffer, originalSampleRate, targetSampleRate) { if (originalSampleRate === targetSampleRate) { From eef7da454a082220b6d106558baf1f36f69aac73 Mon Sep 17 00:00:00 2001 From: GamerBoss101 Date: Sat, 29 Mar 2025 23:54:02 -0400 Subject: [PATCH 08/10] Demo Update 3 --- Backend/server.py | 296 ++++++++++++++++++++++++++++++++++-------- Backend/voice-chat.js | 136 +++++++++++-------- 2 files changed, 320 insertions(+), 112 deletions(-) diff --git a/Backend/server.py b/Backend/server.py index bacf793..b638e99 100644 --- a/Backend/server.py +++ b/Backend/server.py @@ -16,6 +16,28 @@ import gc from collections import deque from threading import Lock +# Add these lines right after your imports +import torch +import os + +# Handle CUDA issues +os.environ["CUDA_VISIBLE_DEVICES"] = "0" # Limit to first GPU only +torch.backends.cudnn.benchmark = True + +# Set CUDA settings to avoid TF32 warnings +torch.backends.cuda.matmul.allow_tf32 = True +torch.backends.cudnn.allow_tf32 = True + +# Set compute type based on available hardware +if torch.cuda.is_available(): + device = "cuda" + compute_type = "float16" # Faster for CUDA +else: + device = "cpu" + compute_type = "int8" # Better for CPU + +print(f"Using device: {device} with compute type: {compute_type}") + # Select device if torch.cuda.is_available(): device = "cuda" @@ -28,9 +50,22 @@ generator = load_csm_1b(device=device) # Initialize WhisperX for ASR print("Loading WhisperX model...") -# Use a smaller model for faster response times -asr_model = whisperx.load_model("medium", device, compute_type="float16") -print("WhisperX model loaded!") +try: + # Try to load a smaller model for faster response times + asr_model = whisperx.load_model("small", device, compute_type=compute_type) + print("WhisperX 'small' model loaded successfully") +except Exception as e: + print(f"Error loading 'small' model: {str(e)}") + try: + # Fall back to tiny model if small fails + asr_model = whisperx.load_model("tiny", device, compute_type=compute_type) + print("WhisperX 'tiny' model loaded as fallback") + except Exception as e2: + print(f"Error loading fallback model: {str(e2)}") + print("Trying CPU model as last resort") + # Last resort - try CPU + asr_model = whisperx.load_model("tiny", "cpu", compute_type="int8") + print("WhisperX loaded on CPU as last resort") # Silence detection parameters SILENCE_THRESHOLD = 0.01 # Adjust based on your audio normalization @@ -53,76 +88,130 @@ active_clients = {} # Map client_id to client context # Helper function to convert audio data def decode_audio_data(audio_data: str) -> torch.Tensor: - """Decode base64 audio data to a torch tensor""" + """Decode base64 audio data to a torch tensor with improved error handling""" try: # Skip empty audio data - if not audio_data: - print("Empty audio data received") + if not audio_data or len(audio_data) < 100: + print("Empty or too short audio data received") return torch.zeros(generator.sample_rate // 2) # 0.5 seconds of silence # Extract the actual base64 content if ',' in audio_data: + # Handle data URL format (data:audio/wav;base64,...) audio_data = audio_data.split(',')[1] # Decode base64 audio data try: binary_data = base64.b64decode(audio_data) print(f"Decoded base64 data: {len(binary_data)} bytes") + + # Check if we have enough data for a valid WAV + if len(binary_data) < 44: # WAV header is 44 bytes + print("Data too small to be a valid WAV file") + return torch.zeros(generator.sample_rate // 2) except Exception as e: print(f"Base64 decoding error: {str(e)}") return torch.zeros(generator.sample_rate // 2) - # Debug: save the raw binary data to examine with external tools + # Save for debugging debug_path = os.path.join(base_dir, "debug_incoming.wav") with open(debug_path, 'wb') as f: f.write(binary_data) - print(f"Saved debug file to {debug_path}") - - # Load audio from binary data + print(f"Saved debug file: {debug_path}") + + # Approach 1: Load directly with torchaudio try: with BytesIO(binary_data) as temp_file: + temp_file.seek(0) # Ensure we're at the start of the buffer audio_tensor, sample_rate = torchaudio.load(temp_file, format="wav") - print(f"Loaded audio: shape={audio_tensor.shape}, sample_rate={sample_rate}Hz") + print(f"Direct loading success: shape={audio_tensor.shape}, rate={sample_rate}Hz") # Check if audio is valid if audio_tensor.numel() == 0 or torch.isnan(audio_tensor).any(): - print("Warning: Empty or invalid audio data detected") - return torch.zeros(generator.sample_rate // 2) + raise ValueError("Empty or invalid audio tensor detected") except Exception as e: - print(f"Audio loading error: {str(e)}") - # Try saving to a temporary file instead of loading from BytesIO + print(f"Direct loading failed: {str(e)}") + + # Approach 2: Try to fix/normalize the WAV data try: - temp_path = os.path.join(base_dir, "temp_incoming.wav") + # Sometimes WAV headers can be malformed, attempt to fix + temp_path = os.path.join(base_dir, "temp_fixing.wav") with open(temp_path, 'wb') as f: f.write(binary_data) - print(f"Trying to load from file: {temp_path}") - audio_tensor, sample_rate = torchaudio.load(temp_path, format="wav") - print(f"Loaded from file: shape={audio_tensor.shape}, sample_rate={sample_rate}Hz") - os.remove(temp_path) + + # Use a simpler numpy approach as backup + import numpy as np + import wave + + try: + with wave.open(temp_path, 'rb') as wf: + n_channels = wf.getnchannels() + sample_width = wf.getsampwidth() + sample_rate = wf.getframerate() + n_frames = wf.getnframes() + + # Read the frames + frames = wf.readframes(n_frames) + print(f"Wave reading: channels={n_channels}, rate={sample_rate}Hz, frames={n_frames}") + + # Convert to numpy and then to torch + if sample_width == 2: # 16-bit audio + data = np.frombuffer(frames, dtype=np.int16).astype(np.float32) / 32768.0 + elif sample_width == 1: # 8-bit audio + data = np.frombuffer(frames, dtype=np.uint8).astype(np.float32) / 128.0 - 1.0 + else: + raise ValueError(f"Unsupported sample width: {sample_width}") + + # Convert to mono if needed + if n_channels > 1: + data = data.reshape(-1, n_channels) + data = data.mean(axis=1) + + # Convert to torch tensor + audio_tensor = torch.from_numpy(data) + print(f"Successfully converted with numpy: shape={audio_tensor.shape}") + except Exception as wave_error: + print(f"Wave processing failed: {str(wave_error)}") + # Try with torchaudio as last resort + audio_tensor, sample_rate = torchaudio.load(temp_path, format="wav") + + # Clean up + if os.path.exists(temp_path): + os.remove(temp_path) except Exception as e2: - print(f"Secondary audio loading error: {str(e2)}") + print(f"All WAV loading methods failed: {str(e2)}") + print("Returning silence as fallback") return torch.zeros(generator.sample_rate // 2) + # Ensure audio is the right shape (mono) + if len(audio_tensor.shape) > 1 and audio_tensor.shape[0] > 1: + audio_tensor = torch.mean(audio_tensor, dim=0) + + # Ensure we have a 1D tensor + audio_tensor = audio_tensor.squeeze() + # Resample if needed if sample_rate != generator.sample_rate: try: print(f"Resampling from {sample_rate}Hz to {generator.sample_rate}Hz") - audio_tensor = torchaudio.functional.resample( - audio_tensor.squeeze(0), + resampler = torchaudio.transforms.Resample( orig_freq=sample_rate, new_freq=generator.sample_rate ) - print(f"Resampled audio shape: {audio_tensor.shape}") + audio_tensor = resampler(audio_tensor) except Exception as e: print(f"Resampling error: {str(e)}") - return torch.zeros(generator.sample_rate // 2) - else: - audio_tensor = audio_tensor.squeeze(0) - - print(f"Final audio tensor shape: {audio_tensor.shape}") + # If resampling fails, just return the original audio + # The model can often handle different sample rates + + # Normalize audio to avoid issues + if torch.abs(audio_tensor).max() > 0: + audio_tensor = audio_tensor / torch.abs(audio_tensor).max() + + print(f"Final audio tensor: shape={audio_tensor.shape}, min={audio_tensor.min().item():.4f}, max={audio_tensor.max().item():.4f}") return audio_tensor except Exception as e: - print(f"Error decoding audio: {str(e)}") + print(f"Unhandled error in decode_audio_data: {str(e)}") # Return a small silent audio segment as fallback return torch.zeros(generator.sample_rate // 2) # 0.5 seconds of silence @@ -143,6 +232,8 @@ def transcribe_audio(audio_tensor: torch.Tensor) -> str: temp_path = os.path.join(base_dir, "temp_audio.wav") torchaudio.save(temp_path, audio_tensor.unsqueeze(0).cpu(), generator.sample_rate) + print(f"Transcribing audio file: {temp_path} (size: {os.path.getsize(temp_path)} bytes)") + # Load and transcribe the audio audio = whisperx.load_audio(temp_path) result = asr_model.transcribe(audio, batch_size=16) @@ -155,11 +246,15 @@ def transcribe_audio(audio_tensor: torch.Tensor) -> str: if result["segments"] and len(result["segments"]) > 0: # Combine all segments transcription = " ".join([segment["text"] for segment in result["segments"]]) + print(f"Transcription successful: '{transcription.strip()}'") return transcription.strip() else: + print("Transcription returned no segments") return "" except Exception as e: print(f"Error in transcription: {str(e)}") + import traceback + traceback.print_exc() if os.path.exists("temp_audio.wav"): os.remove("temp_audio.wav") return "" @@ -385,43 +480,73 @@ def handle_stream_audio(data): # Log the transcription print(f"[{client_id}] Transcribed text: '{transcribed_text}'") - # Add to conversation context + # Handle the transcription result if transcribed_text: + # Add user message to context user_segment = Segment(text=transcribed_text, speaker=speaker_id, audio=full_audio) client['context_segments'].append(user_segment) - # Generate a contextual response - response_text = generate_response(transcribed_text, client['context_segments']) - # Send the transcribed text to client emit('transcription', { 'type': 'transcription', 'text': transcribed_text }) - # Generate audio for the response - audio_tensor = generator.generate( - text=response_text, - speaker=1 if speaker_id == 0 else 0, # Use opposite speaker - context=client['context_segments'], - max_audio_length_ms=10_000, - ) + # Generate a contextual response + response_text = generate_response(transcribed_text, client['context_segments']) + print(f"[{client_id}] Generating audio response: '{response_text}'") - # Add response to context - ai_segment = Segment( - text=response_text, - speaker=1 if speaker_id == 0 else 0, - audio=audio_tensor - ) - client['context_segments'].append(ai_segment) - - # Convert audio to base64 and send back to client - audio_base64 = encode_audio_data(audio_tensor) - emit('audio_response', { - 'type': 'audio_response', - 'text': response_text, - 'audio': audio_base64 + # Let the client know we're processing + emit('processing_status', { + 'type': 'processing_status', + 'status': 'generating_audio', + 'message': 'Generating audio response...' }) + + # Generate audio for the response + try: + # Use a different speaker than the user + ai_speaker_id = 1 if speaker_id == 0 else 0 + + # Start audio generation with streaming (chunk by chunk) + audio_chunks = [] + + # This version tries to stream the audio generation in smaller chunks + # Note: CSM model doesn't natively support incremental generation, + # so we're simulating it here for a more responsive UI experience + + # Generate the full response + audio_tensor = generator.generate( + text=response_text, + speaker=ai_speaker_id, + context=client['context_segments'], + max_audio_length_ms=10_000, + ) + + # Add response to context + ai_segment = Segment( + text=response_text, + speaker=ai_speaker_id, + audio=audio_tensor + ) + client['context_segments'].append(ai_segment) + + # Convert audio to base64 and send back to client + audio_base64 = encode_audio_data(audio_tensor) + emit('audio_response', { + 'type': 'audio_response', + 'text': response_text, + 'audio': audio_base64 + }) + + print(f"[{client_id}] Audio response sent: {len(audio_base64)} bytes") + + except Exception as gen_error: + print(f"Error generating audio response: {str(gen_error)}") + emit('error', { + 'type': 'error', + 'message': "Sorry, there was an error generating the audio response." + }) else: # If transcription failed, send a generic response emit('error', { @@ -437,6 +562,7 @@ def handle_stream_audio(data): # If buffer gets too large without silence, process it anyway elif len(client['streaming_buffer']) >= 30: # ~6 seconds of audio at 5 chunks/sec + print(f"[{client_id}] Processing long audio segment without silence") full_audio = torch.cat(client['streaming_buffer'], dim=0) # Process with WhisperX speech-to-text @@ -453,7 +579,9 @@ def handle_stream_audio(data): 'text': transcribed_text + " (processing continued speech...)" }) - client['streaming_buffer'] = [] + # Keep half of the buffer for context (sliding window approach) + half_point = len(client['streaming_buffer']) // 2 + client['streaming_buffer'] = client['streaming_buffer'][half_point:] except Exception as e: import traceback @@ -497,6 +625,62 @@ def handle_stop_streaming(data): 'status': 'stopped' }) +def stream_audio_to_client(client_id, audio_tensor, text, speaker_id, chunk_size_ms=500): + """Stream audio to client in chunks to simulate real-time generation""" + try: + if client_id not in active_clients: + print(f"Client {client_id} not found for streaming") + return + + # Calculate chunk size in samples + chunk_size = int(generator.sample_rate * chunk_size_ms / 1000) + total_chunks = math.ceil(audio_tensor.size(0) / chunk_size) + + print(f"Streaming audio in {total_chunks} chunks of {chunk_size_ms}ms each") + + # Send initial response with text but no audio yet + socketio.emit('audio_response_start', { + 'type': 'audio_response_start', + 'text': text, + 'total_chunks': total_chunks + }, room=client_id) + + # Stream each chunk + for i in range(total_chunks): + start_idx = i * chunk_size + end_idx = min(start_idx + chunk_size, audio_tensor.size(0)) + + # Extract chunk + chunk = audio_tensor[start_idx:end_idx] + + # Encode chunk + chunk_base64 = encode_audio_data(chunk) + + # Send chunk + socketio.emit('audio_response_chunk', { + 'type': 'audio_response_chunk', + 'chunk_index': i, + 'total_chunks': total_chunks, + 'audio': chunk_base64, + 'is_last': i == total_chunks - 1 + }, room=client_id) + + # Brief pause between chunks to simulate streaming + time.sleep(0.1) + + # Send completion message + socketio.emit('audio_response_complete', { + 'type': 'audio_response_complete', + 'text': text + }, room=client_id) + + print(f"Audio streaming complete: {total_chunks} chunks sent") + + except Exception as e: + print(f"Error streaming audio to client: {str(e)}") + import traceback + traceback.print_exc() + if __name__ == "__main__": print(f"\n{'='*60}") print(f"🔊 Sesame AI Voice Chat Server (Flask Implementation)") diff --git a/Backend/voice-chat.js b/Backend/voice-chat.js index c85da8a..b224b27 100644 --- a/Backend/voice-chat.js +++ b/Backend/voice-chat.js @@ -466,37 +466,27 @@ function sendAudioChunk(audioData, speaker) { return; } - console.log(`Creating WAV from audio data: length=${audioData.length}`); + console.log(`Preparing audio chunk: length=${audioData.length}, speaker=${speaker}`); // Check for NaN or invalid values - let hasNaN = false; - let min = Infinity; - let max = -Infinity; - let sum = 0; - + let hasInvalidValues = false; for (let i = 0; i < audioData.length; i++) { if (isNaN(audioData[i]) || !isFinite(audioData[i])) { - hasNaN = true; + hasInvalidValues = true; console.warn(`Invalid audio value at index ${i}: ${audioData[i]}`); break; } - min = Math.min(min, audioData[i]); - max = Math.max(max, audioData[i]); - sum += audioData[i]; } - if (hasNaN) { - console.warn('Audio data contains NaN or Infinity values. Creating silent audio instead.'); + if (hasInvalidValues) { + console.warn('Audio data contains invalid values. Creating silent audio.'); audioData = new Float32Array(audioData.length).fill(0); - } else { - const avg = sum / audioData.length; - console.log(`Audio stats: min=${min.toFixed(4)}, max=${max.toFixed(4)}, avg=${avg.toFixed(4)}`); } try { - // Create WAV blob with proper format + // Create WAV blob const wavData = createWavBlob(audioData, 24000); - console.log(`WAV blob created: size=${wavData.size} bytes, type=${wavData.type}`); + console.log(`WAV blob created: ${wavData.size} bytes`); const reader = new FileReader(); @@ -504,28 +494,21 @@ function sendAudioChunk(audioData, speaker) { try { // Get base64 data const base64data = reader.result; - console.log(`Base64 data created: length=${base64data.length}`); + console.log(`Base64 data created: ${base64data.length} bytes`); - // Validate the base64 data before sending - if (!base64data || base64data.length < 100) { - console.warn('Generated base64 data is too small or invalid'); - return; - } - - // Send the audio chunk to the server - console.log('Sending audio data to server...'); + // Send to server state.socket.emit('stream_audio', { audio: base64data, speaker: speaker }); - console.log('Audio data sent successfully'); + console.log('Audio chunk sent to server'); } catch (err) { console.error('Error preparing audio data:', err); } }; - reader.onerror = function(err) { - console.error('Error reading audio data:', err); + reader.onerror = function() { + console.error('Error reading audio data as base64'); }; reader.readAsDataURL(wavData); @@ -534,19 +517,20 @@ function sendAudioChunk(audioData, speaker) { } } -// Create WAV blob from audio data with validation +// Create WAV blob from audio data with improved error handling function createWavBlob(audioData, sampleRate) { - // Check if audio data is valid + // Validate input if (!audioData || audioData.length === 0) { - console.warn('Empty audio data received'); - // Return a tiny silent audio snippet instead - audioData = new Float32Array(100).fill(0); + console.warn('Empty audio data provided to createWavBlob'); + audioData = new Float32Array(1024).fill(0); // Create 1024 samples of silence } // Function to convert Float32Array to Int16Array for WAV format function floatTo16BitPCM(output, offset, input) { for (let i = 0; i < input.length; i++, offset += 2) { + // Ensure values are in -1 to 1 range const s = Math.max(-1, Math.min(1, input[i])); + // Convert to 16-bit PCM output.setInt16(offset, s < 0 ? s * 0x8000 : s * 0x7FFF, true); } } @@ -558,40 +542,80 @@ function createWavBlob(audioData, sampleRate) { } } - // Create WAV file with header - function encodeWAV(samples) { - const buffer = new ArrayBuffer(44 + samples.length * 2); + try { + // Create WAV file with header - careful with buffer sizes + const buffer = new ArrayBuffer(44 + audioData.length * 2); const view = new DataView(buffer); - // RIFF chunk descriptor + // RIFF identifier writeString(view, 0, 'RIFF'); - view.setUint32(4, 36 + samples.length * 2, true); + + // File length (will be filled later) + view.setUint32(4, 36 + audioData.length * 2, true); + + // WAVE identifier writeString(view, 8, 'WAVE'); - // fmt sub-chunk + // fmt chunk identifier writeString(view, 12, 'fmt '); + + // fmt chunk length view.setUint32(16, 16, true); - view.setUint16(20, 1, true); // PCM format - view.setUint16(22, 1, true); // Mono channel + + // Sample format (1 is PCM) + view.setUint16(20, 1, true); + + // Mono channel + view.setUint16(22, 1, true); + + // Sample rate view.setUint32(24, sampleRate, true); - view.setUint32(28, sampleRate * 2, true); // Byte rate - view.setUint16(32, 2, true); // Block align - view.setUint16(34, 16, true); // Bits per sample - // data sub-chunk + // Byte rate (sample rate * block align) + view.setUint32(28, sampleRate * 2, true); + + // Block align (channels * bytes per sample) + view.setUint16(32, 2, true); + + // Bits per sample + view.setUint16(34, 16, true); + + // data chunk identifier writeString(view, 36, 'data'); - view.setUint32(40, samples.length * 2, true); - floatTo16BitPCM(view, 44, samples); - return buffer; + // data chunk length + view.setUint32(40, audioData.length * 2, true); + + // Write the PCM samples + floatTo16BitPCM(view, 44, audioData); + + // Create and return blob + return new Blob([view], { type: 'audio/wav' }); + } catch (err) { + console.error('Error in createWavBlob:', err); + + // Create a minimal valid WAV file with silence as fallback + const fallbackSamples = new Float32Array(1024).fill(0); + const fallbackBuffer = new ArrayBuffer(44 + fallbackSamples.length * 2); + const fallbackView = new DataView(fallbackBuffer); + + writeString(fallbackView, 0, 'RIFF'); + fallbackView.setUint32(4, 36 + fallbackSamples.length * 2, true); + writeString(fallbackView, 8, 'WAVE'); + writeString(fallbackView, 12, 'fmt '); + fallbackView.setUint32(16, 16, true); + fallbackView.setUint16(20, 1, true); + fallbackView.setUint16(22, 1, true); + fallbackView.setUint32(24, sampleRate, true); + fallbackView.setUint32(28, sampleRate * 2, true); + fallbackView.setUint16(32, 2, true); + fallbackView.setUint16(34, 16, true); + writeString(fallbackView, 36, 'data'); + fallbackView.setUint32(40, fallbackSamples.length * 2, true); + floatTo16BitPCM(fallbackView, 44, fallbackSamples); + + return new Blob([fallbackView], { type: 'audio/wav' }); } - - // Convert audio data to TypedArray if it's a regular Array - const samples = Array.isArray(audioData) ? new Float32Array(audioData) : audioData; - - // Create WAV blob - const wavBuffer = encodeWAV(samples); - return new Blob([wavBuffer], { type: 'audio/wav' }); } // Draw audio visualizer From 230117a0225b9df857810defbcfa9487a3bf6755 Mon Sep 17 00:00:00 2001 From: GamerBoss101 Date: Sun, 30 Mar 2025 00:14:47 -0400 Subject: [PATCH 09/10] Demo Update 4 --- Backend/server.py | 150 +++++++++++++++++++++++++++++++++------------- 1 file changed, 107 insertions(+), 43 deletions(-) diff --git a/Backend/server.py b/Backend/server.py index b638e99..a6b70a3 100644 --- a/Backend/server.py +++ b/Backend/server.py @@ -16,56 +16,91 @@ import gc from collections import deque from threading import Lock -# Add these lines right after your imports -import torch -import os +# Add this at the top of your file, replacing your current CUDA setup -# Handle CUDA issues -os.environ["CUDA_VISIBLE_DEVICES"] = "0" # Limit to first GPU only -torch.backends.cudnn.benchmark = True - -# Set CUDA settings to avoid TF32 warnings -torch.backends.cuda.matmul.allow_tf32 = True -torch.backends.cudnn.allow_tf32 = True - -# Set compute type based on available hardware -if torch.cuda.is_available(): - device = "cuda" - compute_type = "float16" # Faster for CUDA -else: +# CUDA setup with robust error handling +try: + # Handle CUDA issues + os.environ["CUDA_VISIBLE_DEVICES"] = "0" # Limit to first GPU only + + # Try enabling TF32 precision + try: + torch.backends.cuda.matmul.allow_tf32 = True + torch.backends.cudnn.allow_tf32 = True + except: + pass # Ignore if not supported + + # Check if CUDA is available + if torch.cuda.is_available(): + try: + # Test CUDA functionality + x = torch.rand(10, device="cuda") + y = x + x + del x, y + device = "cuda" + compute_type = "float16" + print("CUDA is fully functional") + except Exception as cuda_error: + print(f"CUDA is available but not working correctly: {str(cuda_error)}") + device = "cpu" + compute_type = "int8" + else: + device = "cpu" + compute_type = "int8" +except Exception as e: + print(f"Error setting up CUDA: {str(e)}") device = "cpu" - compute_type = "int8" # Better for CPU + compute_type = "int8" print(f"Using device: {device} with compute type: {compute_type}") -# Select device -if torch.cuda.is_available(): - device = "cuda" -else: - device = "cpu" -print(f"Using device: {device}") +# Initialize the Sesame CSM model with robust error handling +try: + print(f"Loading Sesame CSM model on {device}...") + generator = load_csm_1b(device=device) + print("Sesame CSM model loaded successfully") +except Exception as model_error: + print(f"Error loading Sesame CSM on {device}: {str(model_error)}") + if device == "cuda": + # Try on CPU as fallback + try: + print("Trying to load Sesame CSM on CPU instead...") + device = "cpu" # Update global device setting + generator = load_csm_1b(device="cpu") + print("Sesame CSM model loaded on CPU successfully") + except Exception as cpu_error: + print(f"Fatal error - could not load Sesame CSM model: {str(cpu_error)}") + raise RuntimeError("Failed to load speech synthesis model") + else: + # Already tried CPU and it failed + raise RuntimeError("Failed to load speech synthesis model on any device") -# Initialize the model -generator = load_csm_1b(device=device) - -# Initialize WhisperX for ASR +# Initialize WhisperX for ASR with robust error handling print("Loading WhisperX model...") try: - # Try to load a smaller model for faster response times - asr_model = whisperx.load_model("small", device, compute_type=compute_type) - print("WhisperX 'small' model loaded successfully") + # First try the smallest model ("tiny") to avoid memory issues + asr_model = whisperx.load_model("tiny", device, compute_type=compute_type) + print("WhisperX 'tiny' model loaded successfully") + + # If tiny worked and we have CUDA, try upgrading to small + if device == "cuda": + try: + asr_model = whisperx.load_model("small", device, compute_type=compute_type) + print("WhisperX 'small' model loaded successfully") + except Exception as upgrade_error: + print(f"Staying with 'tiny' model: {str(upgrade_error)}") except Exception as e: - print(f"Error loading 'small' model: {str(e)}") + print(f"Error loading models on {device}: {str(e)}") + print("Falling back to CPU model") try: - # Fall back to tiny model if small fails - asr_model = whisperx.load_model("tiny", device, compute_type=compute_type) - print("WhisperX 'tiny' model loaded as fallback") - except Exception as e2: - print(f"Error loading fallback model: {str(e2)}") - print("Trying CPU model as last resort") - # Last resort - try CPU + # Force CPU as last resort + device = "cpu" + compute_type = "int8" asr_model = whisperx.load_model("tiny", "cpu", compute_type="int8") print("WhisperX loaded on CPU as last resort") + except Exception as cpu_error: + print(f"Fatal error - could not load any model: {str(cpu_error)}") + raise RuntimeError("No ASR model could be loaded. Please check your CUDA installation.") # Silence detection parameters SILENCE_THRESHOLD = 0.01 # Adjust based on your audio normalization @@ -226,7 +261,7 @@ def encode_audio_data(audio_tensor: torch.Tensor) -> str: def transcribe_audio(audio_tensor: torch.Tensor) -> str: - """Transcribe audio using WhisperX""" + """Transcribe audio using WhisperX with robust error handling""" try: # Save the tensor to a temporary file temp_path = os.path.join(base_dir, "temp_audio.wav") @@ -234,9 +269,38 @@ def transcribe_audio(audio_tensor: torch.Tensor) -> str: print(f"Transcribing audio file: {temp_path} (size: {os.path.getsize(temp_path)} bytes)") - # Load and transcribe the audio - audio = whisperx.load_audio(temp_path) - result = asr_model.transcribe(audio, batch_size=16) + # Load the audio file using whisperx's function + try: + audio = whisperx.load_audio(temp_path) + except Exception as audio_load_error: + print(f"WhisperX load_audio failed: {str(audio_load_error)}") + # Fall back to manual loading + import soundfile as sf + audio, sr = sf.read(temp_path) + if sr != 16000: # WhisperX expects 16kHz audio + from scipy import signal + audio = signal.resample(audio, int(len(audio) * 16000 / sr)) + + # Transcribe with error handling for CUDA issues + try: + # Try with original device + result = asr_model.transcribe(audio, batch_size=8) + except RuntimeError as cuda_error: + if "CUDA" in str(cuda_error) or "libcudnn" in str(cuda_error): + print(f"CUDA error in transcription, falling back to CPU: {str(cuda_error)}") + + # Try to load a CPU model as fallback + try: + global asr_model + # Move model to CPU and try again + asr_model = whisperx.load_model("tiny", "cpu", compute_type="int8") + result = asr_model.transcribe(audio, batch_size=1) + except Exception as e: + print(f"CPU fallback also failed: {str(e)}") + return "I'm having trouble processing audio right now." + else: + # Re-raise if it's not a CUDA error + raise # Clean up if os.path.exists(temp_path): @@ -257,7 +321,7 @@ def transcribe_audio(audio_tensor: torch.Tensor) -> str: traceback.print_exc() if os.path.exists("temp_audio.wav"): os.remove("temp_audio.wav") - return "" + return "I heard something but couldn't understand it." def generate_response(text: str, conversation_history: List[Segment]) -> str: From bb5e0c4765f010d4bd313d1b4d7198e43c764ac5 Mon Sep 17 00:00:00 2001 From: GamerBoss101 Date: Sun, 30 Mar 2025 00:17:39 -0400 Subject: [PATCH 10/10] Demo Fixes 1 --- Backend/server.py | 60 +++++++++++++++++++++++++++++++---------------- 1 file changed, 40 insertions(+), 20 deletions(-) diff --git a/Backend/server.py b/Backend/server.py index a6b70a3..d0dee80 100644 --- a/Backend/server.py +++ b/Backend/server.py @@ -75,32 +75,51 @@ except Exception as model_error: # Already tried CPU and it failed raise RuntimeError("Failed to load speech synthesis model on any device") +# Replace the WhisperX model loading section + # Initialize WhisperX for ASR with robust error handling print("Loading WhisperX model...") +asr_model = None # Initialize to None first to avoid scope issues + try: - # First try the smallest model ("tiny") to avoid memory issues - asr_model = whisperx.load_model("tiny", device, compute_type=compute_type) - print("WhisperX 'tiny' model loaded successfully") + # Always start with the tiny model on CPU for stability + asr_model = whisperx.load_model("tiny", "cpu", compute_type="int8") + print("WhisperX 'tiny' model loaded on CPU successfully") - # If tiny worked and we have CUDA, try upgrading to small + # If CPU works, try CUDA if available if device == "cuda": try: - asr_model = whisperx.load_model("small", device, compute_type=compute_type) - print("WhisperX 'small' model loaded successfully") - except Exception as upgrade_error: - print(f"Staying with 'tiny' model: {str(upgrade_error)}") + print("Trying to load WhisperX on CUDA...") + cuda_model = whisperx.load_model("tiny", "cuda", compute_type="float16") + # Test the model to ensure it works + test_audio = torch.zeros(16000) # 1 second of silence at 16kHz + _ = cuda_model.transcribe(test_audio.numpy(), batch_size=1) + # If we get here, CUDA works + asr_model = cuda_model + print("WhisperX model moved to CUDA successfully") + + # Try to upgrade to small model on CUDA + try: + small_model = whisperx.load_model("small", "cuda", compute_type="float16") + # Test it + _ = small_model.transcribe(test_audio.numpy(), batch_size=1) + asr_model = small_model + print("WhisperX 'small' model loaded on CUDA successfully") + except Exception as upgrade_error: + print(f"Staying with 'tiny' model on CUDA: {str(upgrade_error)}") + except Exception as cuda_error: + print(f"CUDA loading failed, staying with CPU model: {str(cuda_error)}") except Exception as e: - print(f"Error loading models on {device}: {str(e)}") - print("Falling back to CPU model") - try: - # Force CPU as last resort - device = "cpu" - compute_type = "int8" - asr_model = whisperx.load_model("tiny", "cpu", compute_type="int8") - print("WhisperX loaded on CPU as last resort") - except Exception as cpu_error: - print(f"Fatal error - could not load any model: {str(cpu_error)}") - raise RuntimeError("No ASR model could be loaded. Please check your CUDA installation.") + print(f"Error loading WhisperX model: {str(e)}") + # Create a minimal dummy model as last resort + class DummyModel: + def __init__(self): + self.device = "cpu" + def transcribe(self, *args, **kwargs): + return {"segments": [{"text": "Speech recognition currently unavailable."}]} + + asr_model = DummyModel() + print("WARNING: Using dummy transcription model - ASR functionality limited") # Silence detection parameters SILENCE_THRESHOLD = 0.01 # Adjust based on your audio normalization @@ -262,6 +281,8 @@ def encode_audio_data(audio_tensor: torch.Tensor) -> str: def transcribe_audio(audio_tensor: torch.Tensor) -> str: """Transcribe audio using WhisperX with robust error handling""" + global asr_model # Declare global at the beginning of the function + try: # Save the tensor to a temporary file temp_path = os.path.join(base_dir, "temp_audio.wav") @@ -291,7 +312,6 @@ def transcribe_audio(audio_tensor: torch.Tensor) -> str: # Try to load a CPU model as fallback try: - global asr_model # Move model to CPU and try again asr_model = whisperx.load_model("tiny", "cpu", compute_type="int8") result = asr_model.transcribe(audio, batch_size=1)