Music Generation Service

• Generate audio from a natural‑language prompt.
• Produce a mel‑spectrogram image of the generated audio.
• Serve a tiny web UI that works on any browser (including your Samsung Galaxy A16).

Architecture

• FastAPI server (app.py) – loads the pretrained MusicGen model once and exposes a /generate endpoint.
• Model – uses the open‑source facebook/musicgen-medium checkpoint (requires ~4 GB VRAM for GPU, works on CPU albeit slower).
• Audio processing – librosa + matplotlib to create a spectrogram.
• Web UI – static index.html that sends a POST request to /generate and displays the audio player and spectrogram.

Setup

1. Install dependencies
   bash

   1cd music_gen_service 2python -m venv venv 3source venv/bin/activate # on Windows use `venv\Scripts\activate` 4pip install -r requirements.txt
2. Run the server
   uvicorn app:app --host 0.0.0.0 --port 8000
   The service will be reachable at http://localhost:8000 (or the host IP if you expose it).
3. Access from your phone
   • Ensure the machine running the server is reachable on your local network (e.g., both on same Wi‑Fi).
   • Open the address http://<host‑ip>:8000 in the browser on your Samsung Galaxy A16.
   • Use the UI to type a prompt, set duration, and click Generate.

API (usable by OpenClaw)

• audio_base64 – a WAV audio file you can stream (data:audio/wav;base64,...).
• spectrogram_base64 – a PNG image of the mel‑spectrogram.

Docker (optional)

Notes & Tips

• Performance – GPU dramatically speeds up generation. If you have a CUDA‑compatible GPU, change model.to("cpu") to model.to("cuda") in app.py.
• Safety – MusicGen is a general‑purpose model. Avoid prompting for copyrighted or disallowed content.
• Customization – You can swap the model checkpoint (musicgen-small, musicgen-large) by changing the names in the processor and model loading lines.
• Mobile friendliness – The UI uses only standard HTML/JS, so it works on any modern mobile browser without extra plugins.