Update app.py

app.py

@@ -1,46 +1,49 @@
 import streamlit as st
 from transformers import pipeline
+import librosa
+import soundfile as sf
 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")
+# Load the ASR pipeline with the specified model
+pipe = pipeline("automatic-speech-recognition", model="kingabzpro/wav2vec2-large-xls-r-300m-Urdu")
 
-def audio_to_numpy(audio_file):
- # Load the audio file into an AudioSegment object
- audio = AudioSegment.from_file(io.BytesIO(audio_file.read()))
+def load_audio(audio_file):
+ """Load an audio file and convert to the correct format."""
+ audio_bytes = audio_file.read()
+ audio = io.BytesIO(audio_bytes)
  
- # Convert audio to mono and set sample rate to 16000
- audio = audio.set_channels(1).set_frame_rate(16000)
+ # Use librosa to load the audio file
+ audio_np, sr = librosa.load(audio, sr=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
+ return audio_np, sr
 
-def transcribe_audio(audio_file):
- # Convert audio bytes to numpy array
- audio_numpy = audio_to_numpy(audio_file)
+def transcribe_audio(audio_np):
+ """Transcribe the given audio numpy array using the model pipeline."""
+ # Convert the audio numpy array to a format acceptable by the pipeline
+ audio = sf.write(io.BytesIO(), audio_np, 16000, format='wav')
  
  # Transcribe audio
- transcription = pipe(audio_numpy)
+ transcription = pipe(audio)
+ 
  return transcription['text']
 
 # Streamlit UI
-st.title("Speech-to-Text Transcription App")
-st.write("Upload an audio file to transcribe its content into text.")
+st.title("Urdu Speech-to-Text Transcription App")
+st.write("Upload an audio file to transcribe its content into Urdu text.")
 
-uploaded_file = st.file_uploader("Choose an audio file...", type=["wav", "mp3", "flac"])
+uploaded_file = st.file_uploader("Choose an audio file...", type=["wav", "mp3"])
 
 if uploaded_file is not None:
  try:
- with st.spinner("Transcribing..."):
- text = transcribe_audio(uploaded_file)
- st.subheader("Transcription Result:")
- st.write(text)
+ # Load and process the audio file
+ audio_np, sr = load_audio(uploaded_file)
+ 
+ # Transcribe the audio
+ text = transcribe_audio(audio_np)
+ 
+ # Display the transcription result
+ st.subheader("Transcription Result:")
+ st.write(text)
  except Exception as e:
  st.error(f"An error occurred: {e}")