feat: stable optimized backend with large-v3-turbo and loop prevention

app.py

@@ -15,7 +15,7 @@ from pathlib import Path
 from typing import Optional, Dict, Any, List
 from collections import deque
 
-from fastapi import FastAPI, File, UploadFile, HTTPException, Form, WebSocket, WebSocketDisconnect
+from fastapi import FastAPI, File, UploadFile, HTTPException, Form
 from fastapi.responses import Response
 from fastapi.middleware.cors import CORSMiddleware
 from pydantic import BaseModel
@@ -412,72 +412,6 @@ async def transcribe_audio(
         if temp_path and os.path.exists(temp_path):
             os.unlink(temp_path)
 
-# --- WEBSOCKET STREAMING ---
-
-@app.websocket("/api/transcribe/stream")
-async def websocket_transcribe(websocket: WebSocket, language: str = "en"):
-    await websocket.accept()
-    logger.info(f"New WebSocket connection for streaming transcription (Language: {language})")
-    
-    model = load_whisper_model()
-    # Get the initial prompt for the language if available
-    initial_prompt = LANGUAGE_PROMPTS.get(language, "")
-    
-    audio_buffer = bytearray()
-    
-    try:
-        while True:
-            # Receive binary audio data
-            data = await websocket.receive_bytes()
-            if not data:
-                break
-                
-            audio_buffer.extend(data)
-            
-            # Process if buffer is large enough (e.g., > 1 sec at 16k mono = 32000 bytes)
-            if len(audio_buffer) > 48000: # ~1.5 seconds of audio
-                # Write buffer to a temporary file for Whisper
-                with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as tmp:
-                    tmp.write(audio_buffer)
-                    tmp_path = tmp.name
-                
-                try:
-                    # Transcribe the current buffer
-                    segments, info = model.transcribe(
-                        tmp_path,
-                        task="transcribe",
-                        language=language,
-                        initial_prompt=initial_prompt,
-                        vad_filter=True,
-                        vad_parameters=dict(min_silence_duration_ms=400),
-                        condition_on_previous_text=False
-                    )
-                    
-                    full_text = " ".join([s.text.strip() for s in segments]).strip()
-                    
-                    # Send result back
-                    await websocket.send_json({
-                        "text": full_text,
-                        "language": info.language,
-                        "confidence": info.language_probability,
-                        "is_final": False # Streaming is continuous
-                    })
-                    
-                    # Keep the buffer for context (max 30 seconds)
-                    if len(audio_buffer) > 960000: # 30 seconds
-                        audio_buffer = audio_buffer[-480000:] # Keep last 15 seconds
-                        
-                finally:
-                    if os.path.exists(tmp_path):
-                        os.unlink(tmp_path)
-                        
-    except WebSocketDisconnect:
-        logger.info("WebSocket disconnected")
-    except Exception as e:
-        logger.error(f"WebSocket error: {e}")
-    finally:
-        logger.info("Cleaning up WebSocket connection")
-
 # --- SPECIALIZED WHISPER ENDPOINTS ---
 
 @app.post("/api/transcribe/azure", response_model=TranscriptionResponse)