Rm torchaudio, use librosa

app.py

@@ -13,10 +13,11 @@ from fastapi.staticfiles import StaticFiles
 # AI + LLM
 from transformers import WhisperProcessor, WhisperForConditionalGeneration
 import torch
-import torchaudio
+import librosa
 from google import genai
 from google.genai import types
 
+
 ############################################
 # ── Configuration ────────────────────────
 ############################################
@@ -130,8 +131,7 @@ async def voice_transcribe(file: UploadFile = File(...)):  # noqa: B008
         tmp_path = tmp.name
     try:
         # ── 1. Transcribe
-        waveform, sample_rate = torchaudio.load(tmp_path)
-        speech = waveform[0].numpy()  # Convert to numpy for WhisperProcessor
+        speech, sample_rate = librosa.load(tmp_path, sr=16000)
         inputs = processor(speech, sampling_rate=sample_rate, return_tensors="pt")
         input_features = inputs["input_features"].to("cpu")
         generated_ids = model.generate(input_features)

requirements.txt

@@ -7,8 +7,8 @@ python-multipart       # File uploads
 # Voice‑to‑text (Whisper via 🤗 Transformers)
 transformers==4.38.2  # ensure recent enough
 torch
-torchaudio>=2.1.0
 huggingface_hub
+librosa
 
 # Gemini Flash 2.5
 google-genai