stable-audio-3 in c++
A ggml-based C++ implementation of Stable Audio 3 and the Stable Audio Tools model family for cross-platform and embedded use, providing generation and LoRA training CLIs, an embeddable C ABI, and a local browser studio.
sa3.cpp runs Stable Audio 3 and the related Stable Audio Tools models locally using ggml, without depending on the original Python training framework at inference time. It provides command-line tools for generation and for LoRA training, an embeddable C ABI for integrating audio generation into other applications, and a local browser-based studio for both inference and training, served from small local web servers.
Applications built on sa3.cpp outside this repository include a text-to-synth plugin, a VST3 and REAPER extension, and an Ableton Live extension, all using the embedded C ABI, plus an experimental iOS build for on-device LoRA training and inference.
The classic Stable Audio Open and Foundation-1 models are optional, separately built components, downloaded through a Python helper script and run through a dedicated generation binary with a selectable compact or full-precision encoding.