gradio_app.py

import os
import torch
import torchaudio
import gradio as gr
import look2hear.models

# Setup environment and model
os.environ["CUDA_VISIBLE_DEVICES"] = "2"
device = torch.device("cuda") if torch.cuda.is_available() else torch.device("cpu")

model = look2hear.models.TIGERDNR.from_pretrained("JusperLee/TIGER-DnR", cache_dir="cache")
model.to(device)
model.eval()

# Processing function
def separate_audio(audio_file):
    audio, sr = torchaudio.load(audio_file)
    audio = audio.to(device)

    with torch.no_grad():
        all_target_dialog, all_target_effect, all_target_music = model(audio[None])

    # Save outputs
    dialog_path = "dialog_output.wav"
    effect_path = "effect_output.wav"
    music_path = "music_output.wav"

    torchaudio.save(dialog_path, all_target_dialog.cpu(), sr)
    torchaudio.save(effect_path, all_target_effect.cpu(), sr)
    torchaudio.save(music_path, all_target_music.cpu(), sr)

    return dialog_path, effect_path, music_path

# Gradio UI
demo = gr.Interface(
    fn=separate_audio,
    inputs=gr.Audio(type="filepath", label="Upload Audio File"),
    outputs=[
        gr.Audio(label="Dialog", type="filepath"),
        gr.Audio(label="Effects", type="filepath"),
        gr.Audio(label="Music", type="filepath")
    ],
    title="TIGER-DnR Audio Separator",
    description="Upload a mixed audio file to separate it into dialog, effects, and music using the TIGER-DnR model."
)

if __name__ == "__main__":
    demo.launch()