app.py

import gradio as gr
import os
import shutil
import spaces
import sys

# we will clone the repo and install the dependencies
# NOTE: Still fixing bugs, not release, do not try :) !
# os.system('pip install -r qa_mdt/requirements.txt')
# os.system('pip install xformers==0.0.26.post1')
# os.system('pip install torchlibrosa==0.0.9 librosa==0.9.2')
# os.system('pip install -q pytorch_lightning==2.1.3 torchlibrosa==0.0.9 librosa==0.9.2 ftfy==6.1.1 braceexpand')
# os.system('pip install torch==2.3.0+cu121 torchvision==0.18.0+cu121 torchaudio==2.3.0 --index-url https://download.pytorch.org/whl/cu121')

# only then import the necessary modules from qa_mdt
from qa_mdt.pipeline import MOSDiffusionPipeline


pipe = MOSDiffusionPipeline()

# this runs the pipeline with user input and saves the output as 'awesome.wav'
@spaces.GPU(duration=120)
def generate_waveform(description):
    high_quality_description = "high quality " + description
    pipe(high_quality_description)

    generated_file_path = "./awesome.wav"

    # if os.path.exists(generated_file_path):
    #     return generated_file_path
    # else:
    #     return "Error: Failed to generate the waveform."
    if os.path.exists(generated_file_path):
        waveform_video = gr.make_waveform(audio=generated_file_path, bg_color="#000000", bars_color="#00FF00", bar_count=100, bar_width=1.5, animate=True)
        return waveform_video, generated_file_path
    else:
        return "Error: Failed to generate the waveform."


intro = """
# 🎶 OpenMusic: Diffusion That Plays Music 🎧 🎹

Welcome to **OpenMusic**, a next-gen diffusion model designed to generate high-quality music audio from text descriptions! 

Simply enter a few words describing the vibe, and watch as the model generates a unique track for your input. 

Powered by the QA-MDT model, based on the new research paper linked below.

- [GitHub Repo](https://github.com/ivcylc/qa-mdt) by [@changli](https://github.com/ivcylc) 🎓.
- [Paper](https://arxiv.org/pdf/2405.15863) & [Paper Demo](https://qa-mdt.github.io/ )
- [HuggingFace](https://huggingface.co/jadechoghari/qa_mdt) [@jadechoghari](https://github.com/jadechoghari) 🤗.

Note: The music generation process will take 1-2 minutes 🎶
---

"""

# gradio interface
iface = gr.Interface(
    fn=generate_waveform,
    inputs=gr.Textbox(lines=2, placeholder="Enter a music description here..."),
    # outputs=gr.Audio(label="Download the Music 🎼"),
    outputs=[gr.Video(label="Watch the Waveform 🎼"), gr.Audio(label="Download the Music 🎶")],
    description=intro,
    examples=[
        ["A modern synthesizer creating futuristic soundscapes."],
        ["Acoustic ballad with heartfelt lyrics and soft piano."]
        ],
    cache_examples="lazy"
    # cache_examples=True
)

# Launch the Gradio app
if __name__ == "__main__":
    iface.launch()