Gaadwhin/speech_to_text_api

1import whisper2import tempfile3import os4from fastapi import FastAPI, UploadFile, File, HTTPException5from fastapi.middleware.cors import CORSMiddleware6 7app = FastAPI()8 9# ✅ CORS for frontend10app.add_middleware(11    CORSMiddleware,12    allow_origins=["https://godwin015.github.io"],13    allow_credentials=True,14    allow_methods=["*"],15    allow_headers=["*"],16)17 18# IMPORTANT: Change CACHE_DIR to /tmp for Hugging Face Spaces19# Hugging Face Spaces typically restrict write access to /data,20# but /tmp is generally writable for temporary files and model caches.21CACHE_DIR = "/tmp" 22 23# Ensure the cache directory exists.24# This will create /tmp if it doesn't exist, which it usually does.25os.makedirs(CACHE_DIR, exist_ok=True)26 27# ✅ Load Whisper model with custom cache dir28# The model will now download and cache to the writable /tmp directory.29model = whisper.load_model("tiny", download_root=CACHE_DIR)30 31@app.get("/")32def read_root():33    """34    Root endpoint to confirm the backend is running.35    """36    return {"message": "Whisper transcription backend is up."}37 38@app.post("/api/transcribe")39async def transcribe(audio: UploadFile = File(...), language: str = "en"):40    """41    Transcribes an uploaded audio file using the Whisper model.42 43    Args:44        audio (UploadFile): The audio file to transcribe.45        language (str): The language to transcribe in (e.g., "en" for English).46 47    Returns:48        dict: A dictionary containing the transcription text and detected language.49 50    Raises:51        HTTPException: If the uploaded file is not a valid audio file or if an52                       error occurs during transcription.53    """54    # Validate that the uploaded file is an audio file55    if not audio.content_type.startswith("audio/"):56        raise HTTPException(status_code=400, detail="Please upload a valid audio file.")57 58    try:59        # Create a temporary file to save the uploaded audio60        # Using tempfile.NamedTemporaryFile ensures a unique file name61        # and handles cleanup (though we explicitly remove it later).62        # It defaults to the system's temp directory, which is usually /tmp.63        with tempfile.NamedTemporaryFile(delete=False, suffix=".webm") as tmp:64            tmp.write(await audio.read()) # Write the uploaded audio content to the temp file65            tmp_path = tmp.name # Get the path of the temporary file66        67        # Perform the transcription using the Whisper model68        result = model.transcribe(tmp_path, language=language)69        70        # Clean up the temporary audio file after transcription71        os.remove(tmp_path)72        73        # Return the transcription result74        return {75            "transcription": result["text"],76            "language_detected": result.get("language", "unknown")77        }78    except Exception as e:79        # Catch any exceptions during the process and return a 500 error80        # This helps in debugging by providing the error message.81        raise HTTPException(status_code=500, detail=str(e))