Stable Audio 3: Open-Source Latent Diffusion Models for Variable-Length Audio Generation
By
guardienaveugle
11d ago· 2 min readen
75/100
Toasty
Bagelometer↗
Solid neighbourhood-bakery energy. Trustworthy and warm.
Score75Typepress releaseSentimentpositive
Summary
Stable Audio 3 is a family of latent diffusion models (small, medium, large) for variable-length audio generation and editing. The models can generate several minutes of audio, support inpainting for targeted editing, and use a novel semantic-acoustic autoencoder for efficient latent-space generation. They are trained on licensed and Creative Commons data, can generate music and sounds in under 2 seconds on an H200 GPU, and the small and medium model weights are released open-source for consumer-grade hardware.
Key quotes
· 3 pulledOur latent diffusion models operate on top of a novel semantic-acoustic autoencoder that projects audio into a compact latent space, enabling efficient diffusion-based generation while preserving audio fidelity and encouraging semantic structure in the latent.
We release the weights of small and medium, that can run on consumer-grade hardware, together with their training and inference pipeline.
Stable Audio 3 models are trained on licensed and Creative Commons data to generate music and sounds in less than a 2s on an H200 GPU and less than a few seconds on a MacBook Pro M4.
Stable Audio 3 is a family of fast latent diffusion models (small, medium, large) for variable-length audio generation and editing. Since our models can generate several minutes of audio, variable-length generations are key to avoid the cost of producing
