app.py

import os
os.system('pip install gradio==3.34.0')
os.system('pip freeze')
import sys
sys.path.append('.')
import gradio as gr
os.system('pip install -U torchtext==0.8.0')
#os.system('python setup.py install --install-dir .')
from scipy.io import wavfile

# the timestamp to split at (in seconds)
split_at_timestamp = 6

os.system('./separate_scripts/download_checkpoints.sh')

def inference(audio):
    # read the file and get the sample rate and data
    rate, data = wavfile.read(audio.name) 
    
    # get the frame to split at
    split_at_frame = rate * split_at_timestamp
    
    # split
    left_data, right_data = data[:split_at_frame-1], data[split_at_frame:]  # split
    
    # save the result
    wavfile.write('foo_left.wav', rate, left_data)
    os.system("""python bytesep/inference.py --config_yaml=./scripts/4_train/musdb18/configs/vocals-accompaniment,resunet_subbandtime.yaml --checkpoint_path=./downloaded_checkpoints/resunet143_subbtandtime_vocals_8.8dB_350k_steps.pth --audio_path=foo_left.wav --output_path=sep_vocals.mp3""")
    #os.system('./separate_scripts/separate_vocals.sh ' + audio.name + ' "sep_vocals.mp3"')
    os.system("""python bytesep/inference.py --config_yaml=./scripts/4_train/musdb18/configs/accompaniment-vocals,resunet_subbandtime.yaml --checkpoint_path=./downloaded_checkpoints/resunet143_subbtandtime_accompaniment_16.4dB_350k_steps.pth --audio_path=foo_left.wav --output_path=sep_accompaniment.mp3""")
    #os.system('./separate_scripts/separate_accompaniment.sh ' + audio.name + ' "sep_accompaniment.mp3"')
    #os.system('python separate_scripts/separate.py --audio_path=' +audio.name+' --source_type="accompaniment"')
    #os.system('python separate_scripts/separate.py --audio_path=' +audio.name+' --source_type="vocals"')
    return 'sep_vocals.mp3', 'sep_accompaniment.mp3'
title = "Music Source Separation"
description = "Gradio demo for Music Source Separation. To use it, simply add your audio, or click one of the examples to load them. Input Audio is trimmed to 6 seconds to keep inference time fast for each user on cpu. Currently supports .wav files. Read more at the links below."
article = "<p style='text-align: center'><a href='https://arxiv.org/abs/2109.05418'>Decoupling Magnitude and Phase Estimation with Deep ResUNet for Music Source Separation</a> | <a href='https://github.com/bytedance/music_source_separation'>Github Repo</a></p>"

examples = [['example.wav']]
gr.Interface(
    inference, 
    gr.inputs.Audio(type="file", label="Input"), 
    [gr.outputs.Audio(type="file", label="Vocals"),gr.outputs.Audio(type="file", label="Accompaniment")],
    title=title,
    description=description,
    article=article,
    enable_queue=True,
    examples=examples
    ).launch(debug=True)