app.py

import gradio as gr
import tempfile
import torch
import scipy.io.wavfile as wavfile
from transformers import AutoProcessor, MusicgenForConditionalGeneration

# Load model
processor = AutoProcessor.from_pretrained("facebook/musicgen-small")
model = MusicgenForConditionalGeneration.from_pretrained("facebook/musicgen-small")
device = "cuda" if torch.cuda.is_available() else "cpu"
model.to(device)

def generate_music(prompt, duration):
    if not prompt.strip():
        return None, "Please enter a prompt."
    if duration > 40:
        return None, "❌ Duration too long — max allowed is 40 seconds."

    # Prepare inputs
    inputs = processor(text=[prompt], return_tensors="pt").to(device)

    # Approximate duration scaling
    approx_tokens = int(256 * (duration / 8))
    approx_tokens = min(approx_tokens, 2048)

    with torch.no_grad():
        audio = model.generate(**inputs, do_sample=True, guidance_scale=3, max_new_tokens=approx_tokens)

    sr = model.config.audio_encoder.sampling_rate
    audio_arr = audio[0, 0].cpu().numpy()

    # Save temp file
    tmp = tempfile.NamedTemporaryFile(suffix=".wav", delete=False)
    wavfile.write(tmp.name, rate=sr, data=audio_arr)

    return tmp.name, f"✅ Generated {duration}s (approx) of audio!"

with gr.Blocks(title="MusicGen 🎶") as demo:
    gr.Markdown("# 🎵 MusicGen — Text-to-Music Generator (Stable 40s Version)")
    with gr.Row():
        prompt = gr.Textbox(label="🎼 Describe your music", placeholder="e.g. dreamy lo-fi with soft piano")
        duration = gr.Slider(4, 40, value=15, step=1, label="Duration (seconds)")
    btn = gr.Button("Generate 🎧")
    audio_out = gr.Audio(label="🎶 Output", type="filepath")
    msg = gr.Textbox(label="Status", interactive=False)
    btn.click(generate_music, inputs=[prompt, duration], outputs=[audio_out, msg])

demo.launch(share=True)