Fish Audio 发布 Live 3 双工语音模型，针对电话 AI 智能体设计
做电话客服或外呼智能体的开发者可以关注：Fish Audio 的 Live 3 反应低于 200ms，还能分清报号码时的停顿和真正说完，双模型并行架构让抢话问题少了很多。
Fish Audio 发布了 Live 3，一款面向电话场景 AI 智能体的双工语音模型。Live 3 的响应速度低于 200 毫秒，能够区分朗读电话号码中间的停顿和提问结束的语气差异，避免打断用户或错误抢话。
技术上，Live 3 采用双模型架构：主对话模型实时处理语音交互，同时用另一个模型并行准备完整回答，兼顾低延迟和回答质量。这种设计专门解决电话语音场景中回合切换（turn-taking）的难题，例如号码停顿被误判为说完。
双工语音模型是电话外呼、客服智能体的核心组件，与 Retell、Vapi 等语音智能体平台争夺同一市场。对需要构建电话语音智能体的开发者来说，低于 200 毫秒的反应延迟和细粒度的停顿语义理解是选型时的两个关键指标。
Fish Audio has launched Live 3, its latest duplex voice AI model built for AI agents that operate phone calls.
Live 3 can distinguish a pause in the middle of reading a phone number from the end of a question. Live 3’s current reaction speed is under 200 milliseconds and uses a separate model to prepare the full answer in parallel.
This report is categorized under AITOP's "Models" desk, focusing on China's AI industry and domestic foundation model ecosystem. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows and US-China foundation model benchmarks and commercialization comparison. Key entities and technical topics under continuous tracking include Fish Audio, 语音模型, 智能体, 双工语音.
Source attribution: curated from TestingCatalog (@testingcatalog).
