app.py

import gradio as gr
import torch
import torchaudio
import numpy as np
from pathlib import Path
import tempfile
import os

# Check if CUDA is available
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
print(f"Using device: {device}")

def separate_stems(audio_file, model_name="htdemucs"):
    """
    Separate audio stems using Demucs
    """
    if audio_file is None:
        return None, None, None, None, "❌ Please upload an audio file"
    
    try:
        # Import demucs modules
        from demucs.pretrained import get_model
        from demucs.apply import apply_model
        from demucs.audio import save_audio
        
        # Load the model
        model = get_model(model_name)
        model.to(device)
        model.eval()
        
        # Load audio
        wav, sr = torchaudio.load(audio_file)
        
        # Ensure stereo
        if wav.shape[0] == 1:
            wav = wav.repeat(2, 1)
        elif wav.shape[0] > 2:
            wav = wav[:2]
        
        # Resample if necessary
        if sr != model.samplerate:
            resampler = torchaudio.transforms.Resample(sr, model.samplerate)
            wav = resampler(wav)
            sr = model.samplerate
        
        # Move to device
        wav = wav.to(device)
        
        # Apply the model
        with torch.no_grad():
            sources = apply_model(model, wav.unsqueeze(0))
        
        # Get source names
        source_names = model.sources
        
        # Save separated sources
        output_files = {}
        with tempfile.TemporaryDirectory() as temp_dir:
            for i, source in enumerate(source_names):
                output_path = os.path.join(temp_dir, f"{source}.wav")
                save_audio(sources[0, i], output_path, sr)
                
                # Read the saved file for Gradio
                output_files[source] = output_path
        
        # Return the separated stems (assuming 4 stems: drums, bass, other, vocals)
        stems = [None] * 4
        status_msg = f"✅ Successfully separated into {len(source_names)} stems"
        
        for i, source in enumerate(source_names[:4]):  # Limit to 4 for UI
            if source in output_files:
                stems[i] = output_files[source]
        
        return tuple(stems + [status_msg])
        
    except Exception as e:
        error_msg = f"❌ Error during separation: {str(e)}"
        return None, None, None, None, error_msg

def create_hf_interface():
    """Create Hugging Face Spaces compatible interface"""
    
    with gr.Blocks(
        title="🎵 Music Stem Separator",
        theme=gr.themes.Soft(),
    ) as interface:
        
        gr.Markdown("""
        # 🎵 Music Stem Separator
        
        Separate music into individual stems using **Meta's Demucs** model.
        Upload an audio file and get separated tracks for **drums**, **bass**, **other instruments**, and **vocals**.
        
        ⚡ **Powered by Demucs** - State-of-the-art source separation
        """)
        
        with gr.Row():
            with gr.Column():
                # Input
                audio_input = gr.Audio(
                    type="filepath",
                    label="🎼 Upload Music File"
                )
                
                model_choice = gr.Dropdown(
                    choices=[
                        ("HTDemucs (4 stems)", "htdemucs"),
                        ("HTDemucs FT (4 stems)", "htdemucs_ft"), 
                        ("MDX Extra (4 stems)", "mdx_extra")
                    ],
                    value="htdemucs",
                    label="🤖 Model"
                )
                
                separate_btn = gr.Button(
                    "🎛️ Separate Stems", 
                    variant="primary"
                )
            
            with gr.Column():
                gr.Markdown("""
                ### ℹ️ About Stem Separation
                
                **What you'll get:**
                - 🥁 **Drums**: Percussion and rhythm
                - 🎸 **Bass**: Bass lines and low frequencies  
                - 🎹 **Other**: Instruments, synths, effects
                - 🎤 **Vocals**: Lead and backing vocals
                
                **Tips:**
                - Higher quality input = better separation
                - Processing takes 1-3 minutes depending on length
                - Works best with modern pop/rock music
                """)
        
        # Status
        status_output = gr.Textbox(
            label="Status",
            interactive=False
        )
        
        # Output stems
        gr.Markdown("### 🎶 Separated Stems")
        
        with gr.Row():
            drums_output = gr.Audio(
                label="🥁 Drums",
                interactive=False
            )
            bass_output = gr.Audio(
                label="🎸 Bass", 
                interactive=False
            )
        
        with gr.Row():
            other_output = gr.Audio(
                label="🎹 Other",
                interactive=False
            )
            vocals_output = gr.Audio(
                label="🎤 Vocals",
                interactive=False
            )
        
        # Connect the interface  
        separate_btn.click(
            fn=separate_stems,
            inputs=[audio_input, model_choice],
            outputs=[
                drums_output,
                bass_output, 
                other_output,
                vocals_output,
                status_output
            ]
        )
        
        # Examples - simplified for older Gradio
        gr.Markdown("""
        ### 🎵 Instructions
        1. Upload an audio file (MP3, WAV, FLAC)
        2. Choose a separation model  
        3. Click "Separate Stems"
        4. Download individual tracks when ready
        
        **Supported formats**: MP3, WAV, FLAC (max 50MB)
        **Processing time**: 1-5 minutes depending on file length
        """)
        
        gr.Markdown("""
        ---
        **Note**: This space uses Meta's Demucs for stem separation. Processing time depends on audio length and available compute resources.
        
        **Limitations**: 
        - Max file size: ~50MB
        - Processing time: 1-5 minutes
        - Works best with clear, well-produced music
        """)
    
    return interface

# Create and launch the interface
if __name__ == "__main__":
    demo = create_hf_interface()
    demo.launch()