Fish Audio Releases Live 3, Full-Duplex Voice Model
Fish Audio launches Live 3, a full-duplex voice model for voice agents capable of listening while speaking and reacting in just 186 ms, launching in English.
Stock photo for illustration only, not from the actual event
- Fish Audio introduces Live 3, a full-duplex voice model engineered for modern voice agents.
- The system features simultaneous listening and speaking capabilities while filtering out filler sounds from real interruptions.
- It achieves a rapid reaction time starting at just 186 milliseconds during operation.
- English language support is exclusively available at the initial launch phase.
The artificial intelligence and voice technology landscape witnesses a new milestone as Fish Audio officially rolls out Live 3, a cutting-edge full-duplex voice model tailored specifically to power next-generation voice agents with enhanced natural conversational dynamics.
A primary breakthrough of this model is its full-duplex architecture, which empowers the system to listen to user inputs concurrently while it is actively generating and outputting speech, eliminating the awkward turn-taking delays common in legacy conversational systems.
Beyond simultaneous speech processing, Live 3 possesses advanced discernment to distinguish conversational filler sounds, such as acknowledgments or murmurs, from genuine interruptions. This ensures fluid interactions that closely mirror human-to-human communication.
Stock photo for illustration only, not from the actual event
Achieving sub-200-millisecond response times in a full-duplex audio framework marks a significant leap in conversational AI, effectively removing the unnatural latency found in traditional customer service bots. Furthermore, separating backchannel acknowledgments from actual barge-ins prevents the AI from prematurely cutting off its speech when users simply react during listening.
As the rollout begins, the developer has confirmed that the Live 3 model will support English only at the time of its initial release, paving the way for potential multilingual expansions in future updates.
Source: MarkTechPost
Found something wrong in this article? Report an issue with this article
Comments
Leave a Comment