app.py

import streamlit as st
import torchaudio
import speechbrain as sb
from speechbrain.dataio.dataio import read_audio
from IPython.display import Audio
from speechbrain.pretrained import SepformerSeparation as separator
from scipy.io import wavfile
import io

# Load the pretrained model
model = separator.from_hparams(source="speechbrain/sepformer-whamr-enhancement", savedir='pretrained_models/sepformer-whamr-enhancement')

# Define the Streamlit app
def app():
    st.title("Speech Enhancement using Sepformer")

    # Add a file uploader to allow the user to select an audio file
    uploaded_file = st.file_uploader("Choose an audio file", type=["wav"])

    # If an audio file is uploaded, perform speech enhancement and play the results
    if uploaded_file is not None:
        # Load the uploaded audio file
        audio_bytes = uploaded_file.read()
        with open("uploaded_audio.wav", "wb") as f:
            f.write(audio_bytes)
        signal = read_audio("uploaded_audio.wav").squeeze()

        # Perform speech enhancement using the Sepformer model
        enhanced_speech = model.separate_file(path='uploaded_audio.wav')
        enhanced_signal = enhanced_speech[:, :].detach().cpu().squeeze()
        
        enhanced_bytes = io.BytesIO()
        wavfile.write("enhanced_audio.wav", 8000, enhanced_signal.numpy())
        with open("enhanced_audio.wav", "rb") as f:
            enhanced_byte = f.read()
        #st.audio(enhanced_byte, format='audio/wav')
        # Play the original and enhanced audio
        st.audio(audio_bytes, format='audio/wav')
        st.audio(enhanced_byte, format='audio/wav')

# Run the Streamlit app
if __name__ == '__main__':
    app()