Artificial intelligence / news brief
Fish Audio launches Live 3 for full-duplex voice-agent conversations
- Event
- Published

Fish Audio launched Live 3, an English-only speech model that keeps listening while it talks and learns when to acknowledge, wait or yield. It is available through Fish Agent, an API and enterprise on-premises deployment. Fish says short acknowledgments can begin in 186 milliseconds, but it published no benchmark, latency distribution or independent evaluation.
Why it matters
Moving turn timing into the speech model could reduce brittle silence thresholds in phone agents, if the claimed behavior holds outside selected examples.
Sources
- Introducing Live 3: Built for the rhythm of conversation (external site)Fish Audio · primary source ·
- Fish Audio Releases Live 3, a Full-Duplex Voice Model That Reacts in 186 ms (external site)MarkTechPost · coverage ·
- Fish Audio Live 3: 186 ms Full-Duplex Voice, Still Unscored (external site)OrcaRouter · coverage ·