update-4

app.py

@@ -2,6 +2,8 @@ import streamlit as st
 import stable_whisper
 import json
 import torch
+import soundfile as sf
+import io
 
 # Create a dropdown to select the model
 model_name = st.selectbox("Select a model", ["base", "small", "medium", "large", "large-v2"])
@@ -15,10 +17,10 @@ audiofile = st.file_uploader("Upload an audio file", type=["mp3", "wav"])
 # Create a button to run the prediction
 if st.button('Transcribe'):
     if audiofile is not None:
-        # Read the audio file
-        audio_data = torch.from_numpy(audiofile.read())
-        # Convert the audio data to float
-        audio_data = audio_data.float()
+        # Read the audio file into a numpy array
+        audio_data, _ = sf.read(io.BytesIO(audiofile.read()))
+        # Convert the numpy array to a PyTorch tensor and ensure it's float32
+        audio_data = torch.from_numpy(audio_data).float()
         # Transcribe the audio file
         result = model.transcribe(audio_data)
         # Convert the result to JSON and display it

requirements.txt

@@ -1,4 +1,5 @@
 streamlit
 stable-ts
 torch
-numpy
+numpy
+soundfile