lite-start

api/route/convert_from_mp3.py

@@ -6,6 +6,9 @@ from pathlib import Path
 import uuid
 from typing import List
 import asyncio
+import subprocess
+import shutil
+import uuid
 # moviepy is imported lazily inside the conversion function to avoid
 # failing module import at application startup when the package is missing.
 
@@ -85,4 +88,98 @@ async def video_to_mp3(req: VideoToMp3Request):
     return FileResponse(str(dest), filename=dest.name, media_type="audio/mpeg")
 
 
+# Convert an existing MP3 (in data/audio) to WAV and run transcription
+class Mp3ToTextRequest(BaseModel):
+    filename: str
+    model_path: str
+
+
+def _run_ffmpeg(src: Path, dest: Path):
+    # produce mono 16k WAV suitable for Vosk
+    cmd = [
+        "ffmpeg",
+        "-y",
+        "-i",
+        str(src),
+        "-ac",
+        "1",
+        "-ar",
+        "16000",
+        "-vn",
+        "-f",
+        "wav",
+        str(dest),
+    ]
+    subprocess.run(cmd, check=True)
+
+
+@router.post("/mp3-to-text")
+async def mp3_to_text(req: Mp3ToTextRequest):
+    logger = __import__("logging").getLogger("convert.mp3-to-text")
+    try:
+        if shutil.which("ffmpeg") is None:
+            raise HTTPException(status_code=503, detail="ffmpeg is not available in the runtime. Install ffmpeg in the environment/container.")
+    except Exception:
+        raise HTTPException(status_code=503, detail="ffmpeg is not available in the runtime. Install ffmpeg in the environment/container.")
+
+    data_dir = Path(__file__).resolve().parent.parent.parent / "data"
+    audio_dir = data_dir / "audio"
+    src = (audio_dir / req.filename).resolve()
+    try:
+        src.relative_to(audio_dir)
+    except Exception:
+        raise HTTPException(status_code=400, detail="Invalid filename")
+    if not src.exists() or not src.is_file():
+        raise HTTPException(status_code=404, detail="Source file not found")
+
+    audio_dir.mkdir(parents=True, exist_ok=True)
+    dest = audio_dir / f"{src.stem}_{uuid.uuid4().hex}.wav"
+
+    try:
+        logger.info("Starting ffmpeg conversion %s -> %s", src.name, dest.name)
+        await asyncio.get_running_loop().run_in_executor(None, _run_ffmpeg, src, dest)
+        logger.info("ffmpeg conversion finished")
+    except subprocess.CalledProcessError as e:
+        raise HTTPException(status_code=500, detail=f"ffmpeg conversion failed: {e}")
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=f"conversion failed: {e}")
+
+    # Use class-based transcriber
+    try:
+        from api.lib.audio_processing import VoskModelManager
+    except Exception as e:
+        try:
+            dest.unlink()
+        except Exception:
+            pass
+        raise HTTPException(status_code=500, detail=f"Internal import error: {e}")
+
+    try:
+        logger.info("Starting transcription stage for %s", dest.name)
+        # Resolve model path (relative to project root or absolute) and ensure it's inside ./models
+        project_root = Path(__file__).resolve().parent.parent.parent
+        models_dir = project_root / "models"
+        model_path = Path(req.model_path)
+        if not model_path.is_absolute():
+            model_path = project_root / req.model_path
+        model_path = model_path.resolve()
+        try:
+            model_path.relative_to(models_dir)
+        except Exception:
+            raise HTTPException(status_code=400, detail="model_path must point inside the project's models directory")
+        if not model_path.exists():
+            raise HTTPException(status_code=400, detail=f"Model path not found: {model_path}")
+
+        mgr = VoskModelManager(model_path)
+        text = mgr.transcribe_wav(dest)
+        logger.info("Transcription stage finished for %s", dest.name)
+    finally:
+        try:
+            dest.unlink()
+        except Exception:
+            pass
+
+    return {"text": text}
+
+
 # Note: this endpoint uses the bundled ffmpeg provided by `imageio-ffmpeg`.