Fish Audio Raises $52M for Voice AI

Fish Audio provides text-to-speech, voice-cloning, and voice-agent capabilities. SiliconANGLE reports that its platform offers word-level controls for emotion, tone, inflection, and pacing through more than 15,000 natural-language prompts. The company claims that it can clone a voice from a five-second audio sample in less than 15 seconds and supports 83 languages.

Its flagship model is S2.1 Pro. SiliconANGLE reports that, in company-cited listener tests, 67% of listeners preferred Fish Audio outputs over competing systems. That result is a vendor-reported benchmark rather than an independent evaluation, so teams assessing voice quality would need to validate it against their own languages, speaker styles, latency targets, and consent requirements.

Dealroom reports that Fish Audio has more than eight million users and over $21 million in annual recurring revenue. It also reports that the company intends to offer S2.1 Pro free to developers beginning in August and to expand beyond text-to-speech capabilities.

The round is unusually large for a seed-stage voice AI company. Public reporting frames the funding around an ambition to make voice a default interface for AI models, though Fish Audio's future product scope and commercial terms beyond the reported August offer were not detailed in the available material.

For ML practitioners, expressive speech generation increasingly involves more than intelligible text-to-speech. Comparable systems are evaluated on multilingual pronunciation, speaker consistency, emotion controllability, streaming latency, safety controls around voice replication, and deployment economics. Open-source adoption can supply a meaningful developer distribution channel in this market, while production use generally raises separate questions around provenance, authorization, and abuse prevention.