Add /api/transcribe endpoint with Whisper

app/routes.py

@@ -2,10 +2,15 @@
 
 from datetime import datetime, timezone
 
-from fastapi import APIRouter
+from fastapi import APIRouter, File, HTTPException, UploadFile
+
+from app.services.transcriber import transcribe_bytes
 
 router = APIRouter(prefix="/api")
 
+# File size limit: 10 MB
+MAX_UPLOAD_SIZE = 10 * 1024 * 1024
+
 
 @router.get("/health")
 async def health():
@@ -16,3 +21,52 @@ async def health():
         "version": "0.1.0",
         "timestamp": datetime.now(timezone.utc).isoformat(),
     }
+
+
+@router.post("/transcribe")
+async def transcribe_audio(
+    file: UploadFile = File(...),
+    model: str = "base",
+):
+    """Transcribe an audio file using Whisper.
+
+    Supported formats: wav, mp3, m4a, ogg, flac, webm.
+    Model options: tiny, base, small, medium (default: base).
+    """
+    # Validate model
+    valid_models = {"tiny", "base", "small", "medium"}
+    if model not in valid_models:
+        raise HTTPException(
+            status_code=400,
+            detail=f"Invalid model '{model}'. Use: {', '.join(sorted(valid_models))}",
+        )
+
+    # Validate file type
+    allowed = {"audio/wav", "audio/mpeg", "audio/mp4", "audio/x-m4a",
+               "audio/ogg", "audio/flac", "audio/webm", "audio/x-wav"}
+    if file.content_type and file.content_type not in allowed:
+        raise HTTPException(
+            status_code=400,
+            detail=f"Unsupported format: {file.content_type}. Supported: wav, mp3, m4a, ogg, flac, webm",
+        )
+
+    # Read file
+    contents = await file.read()
+    if len(contents) > MAX_UPLOAD_SIZE:
+        raise HTTPException(
+            status_code=413,
+            detail=f"File too large. Max {MAX_UPLOAD_SIZE // (1024*1024)} MB.",
+        )
+    if len(contents) == 0:
+        raise HTTPException(400, detail="Empty file.")
+
+    # Transcribe
+    try:
+        result = transcribe_bytes(contents, model_name=model)
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=f"Transcription failed: {str(e)}")
+
+    return {
+        "filename": file.filename,
+        **result,
+    }

app/services/transcriber.py

@@ -0,0 +1,55 @@
+"""Whisper transcription service — CPU-only, async-ready."""
+
+import io
+import tempfile
+import time
+from pathlib import Path
+
+_model = None
+_model_name = None
+
+
+def _load_model(name: str = "base"):
+    """Lazy-load Whisper model (downloads on first use)."""
+    global _model, _model_name
+    import whisper
+
+    if _model is None or _model_name != name:
+        _model = whisper.load_model(name)
+        _model_name = name
+    return _model
+
+
+def transcribe_bytes(audio_bytes: bytes, model_name: str = "base") -> dict:
+    """Transcribe audio from bytes. Returns {"text": "...", "segments": [...], "language": "..."}"""
+    t0 = time.time()
+
+    model = _load_model(model_name)
+
+    # Write to temp file (whisper needs a file path or numpy array)
+    suffix = ".wav"
+    with tempfile.NamedTemporaryFile(suffix=suffix, delete=False) as tmp:
+        tmp.write(audio_bytes)
+        tmp_path = tmp.name
+
+    try:
+        result = model.transcribe(tmp_path, fp16=False)  # fp16=False for CPU
+    finally:
+        Path(tmp_path).unlink(missing_ok=True)
+
+    elapsed = round(time.time() - t0, 1)
+
+    return {
+        "text": result["text"].strip(),
+        "segments": [
+            {
+                "start": round(seg["start"], 2),
+                "end": round(seg["end"], 2),
+                "text": seg["text"].strip(),
+            }
+            for seg in result.get("segments", [])
+        ],
+        "language": result.get("language", "unknown"),
+        "duration_seconds": elapsed,
+        "model": model_name,
+    }