Voice-AI Startup Fish Audio Lands $52M Seed Round From Coreline Ventures

Fish Audio announced a $52 million seed round led by Coreline Ventures as its emotionally controllable text-to-speech model surpasses 8 million users in its first year.

Fish Audio, developer of a voice model that lets users control emotional delivery inside synthetic speech, has closed a $52 million seed round led by Coreline Ventures. The company announced the investment on July 28, roughly one year after launch, and says it has already attracted more than 8 million builders to its platform.

The core product, branded S2.1 Pro, is an expressive text-to-speech engine that accepts emotion tags — such as [angry], [whispering], [chuckle], or [long pause] — inserted directly into the text input. The model competes on its ability to sustain lifelike prosody across longer passages, with the company’s landing page framing the five-minute mark, not just a five-second clip, as the real test of quality. Fish Audio also offers companion products including voice cloning from a 15‑second sample, speech‑to‑text transcription with speaker diarization and emotion tags, and an end‑to‑end voice‑agent solution.

The $52 million in seed capital will be used to scale the company’s infrastructure and go‑to‑market efforts, funding compute expansion and sales hires to serve the growing demand from enterprise users who embed Fish Audio’s APIs into real‑time voice agents.

Fish Audio has already secured a roster of integration partners that includes Heygen, Retell AI, LiveKit, Telnyx, Sanas, Viggle, and a dozen others, signaling adoption across AI‑video, conversational‑AI, and telecommunications workflows. The company pitches S2.1 Pro for four distinct use cases: video voiceovers, audiobook narration, character‑voice generation for games and animation, and conversational chatbots where sub‑second latency matters.

On the enterprise side, Fish Audio exposes four APIs — text‑to‑speech, speech‑to‑text, voice cloning, and the full voice‑agent stack — and highlights multilingual support across more than 30 languages. The cloning API promises “perfect fidelity” from a 15‑second sample, a speed that puts pressure on incumbents requiring longer reference audio.