Suno AI Bark

Bark is an open source text to audio model released by Suno. Rather than reading a script the way a conventional speech engine does, it generates audio directly from a text prompt, which lets it produce speech alongside music, background noise and simple sound effects, as well as nonverbal sounds such as laughing, sighing and crying. Pretrained checkpoints are published and ready for inference. The repository is aimed at developers and researchers who want generative audio running on their own hardware. It supports multilingual output, ships a voice prompt library for finding a consistent speaker, and documents long form generation and voice consistency techniques in a notebooks section. A smaller version of the model is available where some quality can be traded for speed, and the project documents a speed increase of roughly two times on GPU and ten times on CPU over the original release. Installation, usage and a FAQ are all covered in the README. There is no pricing. Bark was relicensed under the MIT License in May 2023, which makes the model and its pretrained checkpoints free to use commercially. Suno notes that Bark was built for research purposes and is not a conventional text to speech model, so output can deviate from a prompt in unexpected ways, and the company takes no responsibility for what is generated. Suno's text to music products are separate and live on its own website rather than in this repository.