app.py

import subprocess
import sys

# audiocraft's pinned torchaudio<2.1 would drag torch back; install without deps.
subprocess.run(
    [sys.executable, "-m", "pip", "install", "--no-deps", "audiocraft==1.3.0", "demucs"],
    check=True,
)

import spaces
import gradio as gr
import torchaudio
from audiocraft.models import AudioGen
from audiocraft.data.audio import audio_write

model = AudioGen.get_pretrained('facebook/audiogen-medium')


@spaces.GPU
def infer(prompt, duration):
    
    model.set_generation_params(duration=duration)  # generate 5 seconds.
    descriptions = [prompt]
    wav = model.generate(descriptions)  # generates n samples (referring to the number n of prompts in descriptions)

    for idx, one_wav in enumerate(wav):
        # Will save under {idx}.wav, with loudness normalization at -14 db LUFS.
        audio_write(f'{idx}', one_wav.cpu(), model.sample_rate, strategy="loudness", loudness_compressor=True)

    return "0.wav"

css="""
#col-container{
    margin: 0 auto;
    max-width: 640px;
}

footer {
    visibility: hidden;
}
"""

with gr.Blocks(css=css) as demo:
    with gr.Column(elem_id="col-container"):
        gr.HTML("""
        <h2 style="text-align: left;">
            🎵 Generative Songs
        </h2>
        <p style="text-align: center;">   
        </p>
        """)

        prompt_in = gr.Textbox(label="song prompt")
        with gr.Row():
            duration = gr.Slider(label="Duration", minimum=1, maximum=30, step=1, value=5)
            submit_btn = gr.Button("Generate")
        audio_o = gr.Audio(label="Generative Song result")

    submit_btn.click(
        fn=infer,
        inputs=[prompt_in, duration],
        outputs=[audio_o]
    )

# agent5.com song creator

demo.queue().launch()