Stable Audio 3

TL;DR

Stable Audio 3 introduces fast, variable-length audio generation and editing using latent diffusion models with a novel autoencoder, enabling high-quality, efficient audio synthesis on consumer hardware.

Contribution

The paper presents a family of latent diffusion models with a new semantic-acoustic autoencoder and adversarial post-training for improved audio generation and editing.

Findings

Models generate several minutes of audio efficiently.

Autoencoder preserves audio fidelity and semantic structure.

Post-training accelerates inference and enhances quality.

Abstract

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 full-length generations for short sounds. We also support inpainting, enabling targeted audio editing and the continuation of short recordings. Our 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. Finally, we run adversarial post-training to both accelerate inference and improve generation quality, reducing the number of inference steps while improving fidelity and prompt adherence. 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. We release the weights of small and medium, that can run on consumer-grade hardware, together with their training and inference pipeline.