Fish Audio

Fish Audio is a text-to-speech, voice cloning, and speech-to-text platform built around the in-house S2.1 Pro model, with earlier S2 and S1 checkpoints released as open source. The product's headline capability is expressive, emotionally controllable synthesis: prompts can be annotated with tags like [angry], [sad], [excited], [whispering], [laughing], and [pause] to steer prosody, and the same emotion tags are recognised on the recognition side so ingested speech round-trips with its affect intact. Voice cloning is instant-style, requiring only 10-15 seconds of reference audio, and the shared Voice Library exposes more than two million community-contributed voices across 30+ languages that creators can drop into a project without training their own. The typical workflow is to paste a script into the web studio, pick or clone a voice, sprinkle emotion tags where the delivery needs to shift, preview, and export; developers can skip the UI entirely and hit the REST API or SDKs, which offer streaming with low enough latency to drive real-time agents, IVR bots, and character voices in games. Target users are indie video creators and YouTubers who need voiceover without hiring VO talent, audiobook and podcast producers, game and animation studios building character banks, and product teams wiring TTS into chatbots or accessibility features. Fish Audio distinguishes itself from Eleven-tier competitors on price-per-character, an unusually large public voice library, and the fact that its base models are downloadable if you'd rather self-host than pay per second.