Fish Audio Releases Live 3, a Full-Duplex Voice Model That Reacts in 186 ms

What it does

Fish Audio launched Live 3, a full-duplex voice model designed for voice agents that can listen and speak simultaneously. Unlike traditional voice assistants that pause listening when speaking, Live 3 processes input while talking, enabling real-time interaction. It also distinguishes conversational “mm-hmm” sounds from genuine interruptions, reducing false triggers. The model can respond in as little as 186 milliseconds, marking one of the fastest reaction times reported for voice systems. Currently, Live 3 supports English only.

Why it matters

Live 3 cuts latency in voice interactions close to the human conversational gap, which shifts what is possible for voice AI responsiveness. For builders and businesses relying on voice agents, this means smoother, more natural dialogues that feel less robotic and more human. Agents can handle interruptions and overlapping speech without confusion, improving usability in complex or noisy environments. Faster, more seamless exchanges lower frustration, increase engagement, and could accelerate adoption in customer support, smart devices, and voice-driven automation.

Who it is for

Developers designing voice assistants and interactive voice response (IVR) systems will find Live 3 valuable to create more fluid conversations. Businesses deploying voice AI in call centers, device interfaces, or consumer apps benefit from improved reaction speed and interruption handling. Investors and product teams tracking real-time voice AI advancements get a clear signal that latency and speech overlap are active frontiers for differentiation.

The catch

Live 3 is limited to English at launch, which restricts immediate global application and limits testing in multilingual or code-switching contexts. The model’s real-world performance depends on integration quality and background noise conditions, not covered in the announcement. Also, details on resource demands or hardware requirements are unavailable, which affects deployment feasibility for smaller players or edge devices.

What to watch next

See if Fish Audio opens Live 3 to more languages or releases integration tools soon. Watch how quickly competitors close the latency gap or improve interruption detection. Adoption by major voice AI platforms or services will indicate if Live 3’s speed and duplex capabilities become a new baseline for voice models. Finally, monitor user studies showing whether this technical advance significantly improves real-world voice interactions.

AI Quick Briefs Editorial Desk