Tencent Hunyuan Unveils Hy Realtime Voice Model with Tool Calling and Singing

IT之家 November 13 news, Tencent Hunyuan today released a new real-time speech model Hy Realtime. The model has excellent performance in basic dialogue capabilities such as speech understanding, instruction following, and anthropomorphic expression, while also adding Agentic tool calling and singing capabilities.

Hy Realtime has been launched in products such as Yuanbao voice calls and Tencent customer service, and will also be launched on Tencent Cloud in the near future, providing API services externally.

The official statement said that Hy Realtime is still continuously iterating, and the next step will focus on continuous improvement in areas such as duplex interaction, long-term personalized memory, and foreground-background Agent interaction, bringing users a better interaction experience.

The current version can be experienced through the latest version of the Yuanbao App, by selecting the "Chuyu" voice in the voice call feature; some capabilities can also be experienced by visiting the web Demo.

IT之家 attaches the official introduction as follows:

Hy Realtime performs excellently in understanding user intent and instruction following. In real-world communication, user interaction may be slurred, leave sentences half-finished, or change their words midway to correct themselves. The model can combine dialogue context to restore the intent of such incomplete input, with a significant improvement in understanding accuracy. At the same time, the model's multi-turn instruction following and context management capabilities have been further enhanced, and it can still stably understand user intent in longer continuous conversations and maintain smooth dialogue. In addition to basic dialogue functions, Hy Realtime also has tool calling capabilities. It can accurately determine whether to perform a web search based on dialogue context and user intent, unifying search decision-making, tool calling, and answer generation. In addition to the substantial improvement in "listening comprehension," Hy Realtime's spoken language expression capabilities have also been further upgraded. It comprehensively outperforms the previous generation model in all dimensions of sound quality, pronunciation, prosody, and emotion. Changes and progression of emotion within a conversation are more natural, prosody and intonation processing are more diverse, and empathy with users is further enhanced. Relying on deep Co-Design with the Tencent Music TME Lingyin AI team, Hy Realtime adds singing capabilities, integrating natural language song requests and expressive singing into dialogue, moving interaction from "can speak" to "can sing." Users can request songs by combining singer, album, genre, language, release year, and popularity, and can also continue selecting based on prior context. For example, when a user says "play an English folk song from between 2010 and 2020," the model can combine multi-turn context to understand the song request intent, match the song, and sing it. In terms of singing performance, the Hunyuan team has continuously refined expressiveness, naturalness, and musicality, making singing emotions more layered, articulation and breath transitions more natural, and melody and rhythm expression more coordinated. In internal expert A/B Test evaluations, Hy Realtime showed significant advantages in core experience dimensions. ▲ In application scenarios mainly involving Chinese / mixed Chinese-English, Hunyuan Speech has built an experience-oriented real product evaluation system. In the results of internal expert blind evaluations, there has been significant improvement in basic search-and-answer experiences. ▲ On the internally built basic intelligence test set covering multiple dimensions such as mathematics, science, logical reasoning, knowledge Q&A, complex instructions, text understanding, and multi-turn, the Chinese performance in non-thinking mode is impressive. ▲ On multiple public benchmarks in the field of speech interaction, the comprehensive objective performance is at a leading level.

Advertising statement: External redirect links contained in the article (including but not limited to hyperlinks, QR codes, passcodes, and other forms) are used to convey more information and save selection time. The results are for reference only. All articles on IT之家 containing external links include this statement.