Update app.py

app.py

@@ -1,18 +1,34 @@
 import streamlit as st
 from transformers import pipeline
 import torch
+import numpy as np
+from pydub import AudioSegment
 import io
 
 # Load the ASR pipeline with Whisper model
 pipe = pipeline("automatic-speech-recognition", model="openai/whisper-large-v3")
 
+def audio_to_numpy(audio_file):
+    # Load the audio file into an AudioSegment object
+    audio = AudioSegment.from_file(io.BytesIO(audio_file.read()))
+    
+    # Convert audio to mono and set sample rate to 16000
+    audio = audio.set_channels(1).set_frame_rate(16000)
+    
+    # Convert to numpy array
+    samples = np.array(audio.get_array_of_samples())
+    
+    # Normalize the data
+    samples = samples.astype(np.float32) / np.iinfo(audio.sample_width * 8).max
+    
+    return samples
+
 def transcribe_audio(audio_file):
-    # Load audio file
-    audio_bytes = audio_file.read()
-    audio = io.BytesIO(audio_bytes)
+    # Convert audio bytes to numpy array
+    audio_numpy = audio_to_numpy(audio_file)
     
     # Transcribe audio
-    transcription = pipe(audio)
+    transcription = pipe(audio_numpy)
     return transcription['text']
 
 # Streamlit UI
@@ -22,7 +38,10 @@ st.write("Upload an audio file to transcribe its content into text.")
 uploaded_file = st.file_uploader("Choose an audio file...", type=["wav", "mp3", "flac"])
 
 if uploaded_file is not None:
-    with st.spinner("Transcribing..."):
-        text = transcribe_audio(uploaded_file)
-        st.subheader("Transcription Result:")
-        st.write(text)
+    try:
+        with st.spinner("Transcribing..."):
+            text = transcribe_audio(uploaded_file)
+            st.subheader("Transcription Result:")
+            st.write(text)
+    except Exception as e:
+        st.error(f"An error occurred: {e}")