AmbientGen 🌧️

A text-to-ambient-sound generator powered by AudioLDM2

Type a description like “gentle rain on a window with distant thunder” and get a generated ambient soundscape.

🚀 Try it

Live Demo on Hugging Face Spaces →

🎯 What is this?

This is a learning-in-public project documenting my journey into Generative AI for Audio. I’m building a text-to-ambient-sound app while studying the underlying models, papers, and techniques.

Features:

• 🎧 Quick Generate — 8 hand-crafted ambient presets, one click to generate
• 🎛️ Layer Mixer — generate up to 3 sound layers and mix them with volume control
• ✍️ Custom — write your own prompts with adjustable guidance scale and inference steps

📝 Blog

Follow the build process on the project blog:

1. What is AudioLDM2 and why I’m using it
2. First Sounds — What AudioLDM2 Can and Can’t Do
3. Prompt Engineering for Audio — What Actually Works
4. Building AmbientGen — From Notebook to Product

📚 Reading List

A curated list of papers and resources on generative AI for audio and music → Papers & Resources

🛠️ Tech Stack

• Model: AudioLDM2 via HuggingFace Diffusers
• Interface: Gradio
• Deploy: Hugging Face Spaces (ZeroGPU)
• Blog: GitHub Pages + Jekyll

Project Structure

ambientgen/
├── blog/           # Blog posts (Markdown)
├── app/            # Gradio application code
├── experiments/    # Colab notebooks & experiment logs
├── docs/           # Papers reading list & resources
└── README.md

License

MIT