Tencent unveils hy realtime, a new voice model capable of singing and tool integration

Tencent's Hunyuan has released Hy Realtime, a sophisticated new real-time voice model. This model demonstrates superior performance in voice comprehension and instruction following, while also introducing advanced capabilities such as Agentic tool invocation and singing. The technology is already deployed in products like Yuanbao voice calls and Tencent Customer Service, with plans to soon offer API services via Tencent Cloud. Hy Realtime significantly enhances conversational fluidity by accurately restoring user intent even from incomplete or self-corrected speech inputs, improving accuracy in multi-turn dialogues. Beyond comprehension, the model's expressive capabilities have been upgraded across audio quality, pronunciation, rhythm, and emotion, leading to more natural emotional progression during conversations. A key addition, developed in co-design with Tencent Music TME's Lingyin AI team, is the singing feature. Users can request songs based on parameters like artist, album, genre, language, year of release, and popularity, with the model capable of handling multi-turn requests. Internal expert A/B tests indicate Hy Realtime shows marked advantages in core experience dimensions.