app.py

import streamlit as st
from transformers import pipeline
import librosa
import soundfile as sf
import numpy as np
import io

# Load the ASR pipeline with the specified model
pipe = pipeline("automatic-speech-recognition", model="kingabzpro/wav2vec2-large-xls-r-300m-Urdu")

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)
 
 # Use librosa to load the audio file
 audio_np, sr = librosa.load(audio, sr=16000)
 
 return audio_np, sr

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)
 
 return transcription['text']

# Streamlit UI
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"])

if uploaded_file is not None:
 try:
 # 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}")