Update api.py

api.py

@@ -15,6 +15,7 @@ from fastapi.responses import FileResponse, JSONResponse
 from fastapi.staticfiles import StaticFiles
 from huggingface_hub import HfApi, create_repo, upload_file
 from transformers import WhisperForConditionalGeneration, WhisperProcessor
+import librosa
 
 # ── Env ────────────────────────────────────────────────────────────────────
 load_dotenv()
@@ -79,12 +80,10 @@ app = FastAPI(title="Kasanoma ASR", version="2.0.0")
 _csv_lock = threading.Lock()
 
 # ── Helpers ────────────────────────────────────────────────────────────────
-
 def audio_duration(path: Path) -> float:
-    """Return audio duration in seconds; falls back to 0.0 on error."""
     try:
-        data, sr = sf.read(str(path), always_2d=False)
-        return round(len(data) / sr, 2)
+        y, sr = librosa.load(str(path), sr=None, mono=True)
+        return round(len(y) / sr, 2)
     except Exception:
         return 0.0
 
@@ -140,8 +139,11 @@ def transcribe_path(audio_path: Path) -> tuple[str, str]:
     """
     model, processor, device = get_model()
 
-    # Read + convert
-    audio_data, sampling_rate = sf.read(str(audio_path))
+    audio_data, sampling_rate = librosa.load(str(audio_path), sr=16000, mono=True)
+
+    # Normalise
+    if np.max(np.abs(audio_data)) > 0:
+        audio_data = audio_data / np.max(np.abs(audio_data))
 
     # Stereo → mono
     if len(audio_data.shape) > 1: