Update app.py

app.py

@@ -7,27 +7,30 @@ from fastapi.middleware.cors import CORSMiddleware
 from fastapi.responses import JSONResponse
 from pydantic import BaseModel
 
-# ---- Writable caches for Spaces (fixes PermissionError: '/.cache') ----
-os.environ.setdefault("HF_HOME", "/tmp/hf")
-os.environ.setdefault("HUGGINGFACE_HUB_CACHE", "/tmp/hf")
-os.environ.setdefault("XDG_CACHE_HOME", "/tmp/.cache")
-os.makedirs(os.environ["HF_HOME"], exist_ok=True)
-os.makedirs(os.environ["XDG_CACHE_HOME"], exist_ok=True)
-
-# ---- Transcription (CPU) ----
-from faster_whisper import WhisperModel
+# ===== Writable caches (persist across restarts) =====
+os.environ.setdefault("HF_HOME", "/data/hf")
+os.environ.setdefault("HUGGINGFACE_HUB_CACHE", "/data/hf")
+os.environ.setdefault("XDG_CACHE_HOME", "/data/.cache")
+for p in ("/data", "/data/hf", "/data/.cache", "/tmp"):
+    try:
+        os.makedirs(p, exist_ok=True)
+        os.chmod(p, 0o777)
+    except Exception:
+        pass
 
-MODEL_NAME = os.environ.get("WHISPER_MODEL", "tiny.en")  # fast & CPU-friendly
+# ===== Transcriber (CPU) =====
+from faster_whisper import WhisperModel
+MODEL_NAME = os.environ.get("WHISPER_MODEL", "tiny.en")
 
-app = FastAPI(title="Nuvia Free Transcriber", version="1.1.0")
+app = FastAPI(title="Nuvia Free Transcriber", version="1.2.0")
 app.add_middleware(
     CORSMiddleware,
     allow_origins=["*"], allow_credentials=True,
     allow_methods=["*"], allow_headers=["*"],
 )
 
-# Load model once at startup
-model = WhisperModel(MODEL_NAME, device="cpu", compute_type="int8")
+# Load from /data to avoid runtime downloads
+model = WhisperModel(MODEL_NAME, device="cpu", compute_type="int8", download_root="/data/hf")
 
 class HealthOut(BaseModel):
     ok: bool
@@ -61,13 +64,13 @@ def health():
 @app.post("/transcribe", response_model=TranscribeOut)
 async def transcribe(file: UploadFile = File(...)):
     # Save upload
-    raw = await file.read()
     ext = os.path.splitext(file.filename or "")[1].lower() or ".bin"
     tmp_in = f"/tmp/in{ext}"
+    raw = await file.read()
     with open(tmp_in, "wb") as f:
         f.write(raw)
 
-    # Convert to mono 16k wav for robust decode
+    # Convert to mono 16k WAV
     tmp_wav = "/tmp/in.wav"
     try:
         subprocess.check_call(
@@ -75,13 +78,13 @@ async def transcribe(file: UploadFile = File(...)):
             stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL
         )
     except subprocess.CalledProcessError:
-        return JSONResponse(status_code=400, content={"error": "ffmpeg failed to decode the audio"})
+        return JSONResponse(status_code=400, content={"error": "ffmpeg failed to decode audio"})
 
     duration = ffprobe_duration(tmp_wav)
 
     # Transcribe
     segments, _ = model.transcribe(tmp_wav, language="en")
-    text = "".join([seg.text for seg in segments]).strip()
+    text = "".join(seg.text for seg in segments).strip()
 
     wpm = None
     if duration and duration > 0: