h2

app.py

@@ -3,6 +3,7 @@ import stable_whisper
 import json
 import torch
 import soundfile as sf
+import librosa
 from io import BytesIO
 
 # Create a dropdown to select the model
@@ -20,9 +21,13 @@ audiofile = st.file_uploader("Upload an audio file", type=["mp3", "wav"])
 if st.button('Transcribe'):
     if audiofile is not None:
         # Read the audio file into a numpy array
-        audio_data, _ = sf.read(BytesIO(audiofile.read()))
+        audio_data, sample_rate = sf.read(BytesIO(audiofile.read()))
+        # Resample the audio data if necessary
+        expected_sample_rate = 16000  # replace with the sample rate expected by the model
+        if sample_rate != expected_sample_rate:
+            audio_data = librosa.resample(audio_data, sample_rate, expected_sample_rate)
         # Convert the audio data to float
-        audio_data = torch.from_numpy(audio_data)
+        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

@@ -2,4 +2,5 @@ streamlit
 stable-ts
 torch
 numpy
-soundfile
+soundfile
+librosa