Fish Audio

Fish Audio topped the TTS-Arena2 independent benchmark in 2026 and won a blind user-preference study against ElevenLabs and others. Its S2 model clones any voice from a 15-second sample across 80+ languages with 50+ fine-grained emotion controls ([excited], [whispering], [sad], [laughing]) that outperform most competitors in expressiveness. Community library of 2M+ user-shared voice models. API pricing at ~$15/1M characters — roughly 80% cheaper than ElevenLabs at comparable quality. Also includes STT, SFX generation, and vocal removal. Free plan is personal use only; commercial use requires paid tier.