Fish Audio is a voice AI service made by Hanabi AI Inc. Its main product is text-to-speech (TTS), software that turns written text into spoken audio. It also offers voice cloning, which builds a synthetic copy of a voice from a short recording, and speech-to-text transcription.
A developer sends text to the Fish Audio API over REST or a WebSocket stream and receives audio, for example an MP3 file; official SDKs exist for Python and TypeScript. Tags written inside the text, such as [whispering] or [laughing], control emotion and delivery. The platform hosts over 2,000,000 voices uploaded by users, clones a voice from 15 to 30 seconds of audio, and speaks more than 30 languages. Its own models (fish-speech, S1 and S2) are also published as open weights, model files that others can download, under a paid commercial license; running them on a company's own servers is an Enterprise offer.
The API is pay per use, billed by the amount of text, with free credits to start. A web app for creators is sold as a monthly or yearly subscription with a free tier.