In the catalog since 19.08.2026
Fish Audio
1index_Hanabi AIUSA
Speech synthesis from Hanabi AI: voices text with the S2.1 Pro model, clones a voice from a short sample, and marks emotions with tags directly in the text. The product is built on open Fish Speech models.
Facts
- Legal entity — Hanabi AI Inc., a Delaware corporation, at 1111B S Governors Ave STE 48109, Dover, DE 19904; arbitration — in San Francisco County under California law (fish.audio/terms, as of 19.08.2026). Fish Audio is the company's product platform, and OpenAudio is its research lab. Founder and CEO — Shijia Liao, author of the open voice models So-VITS-SVC, GPT-SoVITS, and Bert-VITS2.
- The OpenAudio S1 model was introduced on 03.06.2025: trained on more than 2 million hours of audio, it supports 13 languages, ranked first in TTS-Arena, and achieved 0.8% WER and 0.4% CER on Seed TTS Eval (fish.audio/blog/introducing-s1).
- API lineup as of 19.08.2026: s2.1-pro, the recommended production model; s2.1-pro-free, the same model for 0 $ for testing and prototypes without latency guarantees; s2-pro and s1, previous generations (docs.fish.audio).
- Open source: the fishaudio/fish-speech repository on GitHub - 32,264 stars, last commit 03.08.2026 (GitHub API, snapshot 19.08.2026). The product offering on the same date: speech synthesis, transcription, voice cloning from a 15-second sample, Story Studio for audiobooks, voice agents, and a library of more than 2,000,000 voices.
Full profile in the works: a “worth it or not” verdict, pricing, access from Russia, and comparisons with nearby alternatives will appear as the card factory progresses.