Tencent launches Tenlai speech model family with ASR, TTS and live translation
Tencent announced on October 29 at its Global Digital Ecosystem Conference that it is releasing the Tenlai speech-model family: Tenlai-ASR v1.0 for speech understanding, Tenlai-TTS v1.0 for speech generation, and Tenlai-LiveTranslate v1.0 for real-time cross-language communication. The models are already running across Tencent Meeting, Yuanbao, Sogou Input, WeCom, ima, QQ Browser, Tencent Maps and QQ Music, which Tencent says have cumulatively served hundreds of millions of users, and the family is now generally available to enterprises and developers through Tencent Cloud. Vendor-reported evaluations include a CpWER 30% below an industry benchmark, diarization error rate reduced from 24% to under 10%, voiceprint matching accuracy of 97%, and a first-token latency of 3-4 seconds for live translation.
Tencent is packaging speech understanding, generation and simultaneous interpretation into one commercially available family, giving developers access to the same stack Tencent already ships inside widely used Chinese consumer apps.
10月29日，在腾讯全球数字生态大会上，腾讯正式发布天籁语音大模型Tenlai，包括天籁语音理解大模型Tenlai-ASR v1.0、天籁语音生成大模型Tenlai-TTS v1.0、天籁同声传译大模型Tenlai-LiveTranslate v1.0。
三大模型分别面向语音理解、语音生成和实时跨语言交流，并已在腾讯会议、元宝、搜狗输入法、企业微信、ima、QQ浏览器、腾讯地图、QQ音乐等多款腾讯 AI 产品上应用，累计服务数亿用户。目前，天籁语音大模型Tenlai已正式通过腾讯云对外开放，为企业与开发者提供领先语音技术支撑。
语音理解：能力全面升级，让Agent真正读懂每一场会
天籁语音理解大模型Tenlai-ASR v1.0，能够精准实现多讲话人区分，多模态上下文理解能力、实时音频流和超长音视频转录识别能力大幅提升，进一步推动沟通交流上下文被完整、准确地沉淀，为大模型、Agent快速理解和调用。
#Tencent#Tenlai-ASR v1.0#Tenlai-TTS v1.0#Tenlai-LiveTranslate v1.0#Tencent Cloud#Tencent Meeting#Yuanbao#WeCom#Tencent Maps#QQ Music#Sogou Input