gradio_app.py

import os
import uuid
import torch
import torchaudio
import torchaudio.transforms as T
import soundfile as sf
import gradio as gr
import spaces
import look2hear.models

# Setup device
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

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

sep_model = look2hear.models.TIGER.from_pretrained("JusperLee/TIGER-speech", cache_dir="cache")
sep_model.to(device).eval()

TARGET_SR = 16000
MAX_SPEAKERS = 4

# --- DnR Function ---
@spaces.GPU()
def separate_dnr(audio_file):
    audio, sr = torchaudio.load(audio_file)
    audio = audio.to(device)

    with torch.no_grad():
        dialog, effect, music = dnr_model(audio[None])

    # Unique output folder
    session_id = uuid.uuid4().hex[:8]
    output_dir = os.path.join("output_dnr", session_id)
    os.makedirs(output_dir, exist_ok=True)

    dialog_path = os.path.join(output_dir, "dialog.wav")
    effect_path = os.path.join(output_dir, "effect.wav")
    music_path = os.path.join(output_dir, "music.wav")

    torchaudio.save(dialog_path, dialog.cpu(), sr)
    torchaudio.save(effect_path, effect.cpu(), sr)
    torchaudio.save(music_path, music.cpu(), sr)

    return dialog_path, effect_path, music_path

# --- Speaker Separation Function ---
@spaces.GPU()
def separate_speakers(audio_path):
    waveform, original_sr = torchaudio.load(audio_path)
    if original_sr != TARGET_SR:
        waveform = T.Resample(orig_freq=original_sr, new_freq=TARGET_SR)(waveform)

    if waveform.dim() == 1:
        waveform = waveform.unsqueeze(0)
    audio_input = waveform.unsqueeze(0).to(device)

    with torch.no_grad():
        ests_speech = sep_model(audio_input)

    ests_speech = ests_speech.squeeze(0)

    # Unique output folder
    session_id = uuid.uuid4().hex[:8]
    output_dir = os.path.join("output_sep", session_id)
    os.makedirs(output_dir, exist_ok=True)

    output_files = []
    for i in range(ests_speech.shape[0]):
        path = os.path.join(output_dir, f"speaker_{i+1}.wav")
        audio_np = ests_speech[i].cpu().numpy()
        sf.write(path, audio_np.T, TARGET_SR)  # Transpose only if shape is [T, C], usually not needed
        output_files.append(path)

    updates = []
    for i in range(MAX_SPEAKERS):
        if i < len(output_files):
            updates.append(gr.update(value=output_files[i], visible=True, label=f"Speaker {i+1}"))
        else:
            updates.append(gr.update(value=None, visible=False))
    return updates

# --- Gradio App ---
with gr.Blocks() as demo:
    gr.Markdown("# TIGER: Time-frequency Interleaved Gain Extraction and Reconstruction for Efficient Speech Separation")
    gr.Markdown("TIGER is a lightweight model for speech separation which effectively extracts key acoustic features through frequency band-split, multi-scale and full-frequency-frame modeling.")

    gr.HTML("""
            <div style="display:flex;column-gap:4px;">
                <a href="https://cslikai.cn/TIGER/">
                    <img src='https://img.shields.io/badge/Project-Page-green'>
                </a> 
    			
                <a href="https://huggingface.co/spaces/fffiloni/TIGER-audio-extraction?duplicate=true">
    				<img src="https://huggingface.co/datasets/huggingface/badges/resolve/main/duplicate-this-space-sm.svg" alt="Duplicate this Space">
    			</a>	
            </div>
        """)
    with gr.Tabs():
        # --- Tab 1: DnR ---
        with gr.Tab("Dialog/Effects/Music Separation (DnR)"):
            gr.Markdown("### Separate Dialog, Effects, and Music from Mixed Audio")

            dnr_input = gr.Audio(type="filepath", label="Upload Audio File")
            dnr_button = gr.Button("Separate Audio")

            gr.Examples(
                examples = ["./test/test_mixture_466.wav"],
                inputs = dnr_input
            )

            dnr_output_dialog = gr.Audio(label="Dialog", type="filepath")
            dnr_output_effect = gr.Audio(label="Effects", type="filepath")
            dnr_output_music = gr.Audio(label="Music", type="filepath")

            dnr_button.click(
                fn=separate_dnr,
                inputs=dnr_input,
                outputs=[dnr_output_dialog, dnr_output_effect, dnr_output_music]
            )

        # --- Tab 2: Speaker Separation ---
        with gr.Tab("Speaker Separation"):
            gr.Markdown("### Separate Individual Speakers from Mixed Speech")

            sep_input = gr.Audio(type="filepath", label="Upload Speech Audio")
            sep_button = gr.Button("Separate Speakers")

            gr.Examples(
                examples = ["./test/mix.wav"],
                inputs = sep_input
            )

            gr.Markdown("#### Separated Speakers")
            sep_outputs = []
            for i in range(MAX_SPEAKERS):
                sep_outputs.append(gr.Audio(label=f"Speaker {i+1}", visible=(i == 0), interactive=False))

            sep_button.click(
                fn=separate_speakers,
                inputs=sep_input,
                outputs=sep_outputs
            )

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