Fish Audio

Fish Speech S2

An open-source (Apache 2.0) zero-shot TTS and voice-cloning model that clones any voice from a 10-second audio sample with sub-300ms streaming latency. Supports 13 languages including Chinese, English, and Japanese with cross-lingual zero-shot output. Ranked #1 on TTS-Arena in March 2026 blind listening tests, outperforming ElevenLabs V3 in A/B testing. Available via API at ~$15 per million characters — approximately 6x cheaper than ElevenLabs at scale.