Fish Audio's Live 3 voice model listens and speaks at once

single source· 1 articles · confidence: low · first seen 2026-11-11 04:43 UTC

What this means for you

Nothing to act on yet — there is no pricing, no evaluation and no date beyond English-only availability, and the 186 ms is Fish Audio's own number. If you are choosing a realtime voice stack, the question to put to any vendor is how it tells a backchannel from an interruption, on your audio.

Fish Audio has released Live 3, a voice model for voice agents that listens while it speaks rather than waiting for the user to finish — full-duplex, meaning the channel carries both directions at once. The company says it can separate backchannel noises such as "mm-hmm" from a genuine interruption, and react in as little as 186 milliseconds. It is English-only at launch. Fish Audio has published no price, no availability date and no independent evaluation; the latency figure is the company's own, with no test date or harness attached.

Key facts

  • ·Fish Audio released Live 3, a full-duplex voice model intended for voice agents. source
  • ·Fish Audio claims a reaction latency of as little as 186 milliseconds. source
  • ·The model is English-only at launch. source
  • ·The model listens while speaking and is claimed to distinguish backchannel sounds from real interruptions. source
  • ·No evaluation date, harness or independent test is given for the latency figure. source

What the sources say

  • MarkTechPost — Brief release note on the latency claim, interruption handling and English-only availability.

Sources

The original reporting. Follow these — they did the work.

← the wire