Update app.py

app.py

@@ -6,26 +6,46 @@ import torchaudio
 import numpy as np
 import io
 
+app = FastAPI()
+import torch
+import torchaudio
+from transformers import pipeline
+from fastapi import FastAPI, UploadFile, File
+
 app = FastAPI()
 
-# Load model once at startup
-asr = pipeline("automatic-speech-recognition", model="openai/whisper-tiny")
+asr = pipeline("automatic-speech-recognition", model="openai/whisper-base")
 
 @app.post("/predict")
 async def predict(file: UploadFile = File(...)):
-    try:
-        # Read file bytes
-        audio_bytes = await file.read()
-        audio, sr = torchaudio.load(io.BytesIO(audio_bytes))
-        # Convert to float32 if necessary
-        if audio.dtype != torch.float32:
-            audio = audio.to(torch.float32)
-        # Convert to numpy array
-        np_audio = audio.squeeze().numpy()
-        # Run inference
-        result = asr({"array": np_audio, "sampling_rate": sr})
-        text = result["text"]
-        return JSONResponse({"text": text})
-    except Exception as e:
-        print("🔥 Error during transcription:", e)
-        return JSONResponse({"error": str(e)}, status_code=500)
+    audio_path = "/tmp/input.wav"
+    with open(audio_path, "wb") as f:
+        f.write(await file.read())
+
+    # Load audio safely (no torchcodec)
+    waveform, sr = torchaudio.load(audio_path)
+    waveform = waveform.to(torch.float32)  # force float
+    result = asr({"array": waveform[0].numpy(), "sampling_rate": sr})
+    return {"text": result["text"]}
+
+# # Load model once at startup
+# asr = pipeline("automatic-speech-recognition", model="openai/whisper-tiny")
+
+# @app.post("/predict")
+# async def predict(file: UploadFile = File(...)):
+#     try:
+#         # Read file bytes
+#         audio_bytes = await file.read()
+#         audio, sr = torchaudio.load(io.BytesIO(audio_bytes))
+#         # Convert to float32 if necessary
+#         if audio.dtype != torch.float32:
+#             audio = audio.to(torch.float32)
+#         # Convert to numpy array
+#         np_audio = audio.squeeze().numpy()
+#         # Run inference
+#         result = asr({"array": np_audio, "sampling_rate": sr})
+#         text = result["text"]
+#         return JSONResponse({"text": text})
+#     except Exception as e:
+#         print("🔥 Error during transcription:", e)
+#         return JSONResponse({"error": str(e)}, status_code=500)