Tencent Hunyuan has released the Hy Realtime real-time speech model, which has been integrated into Yuanbao and Tencent Customer Service.

Article

Tencent Hyunyuan launches the real-time voice model Hy Realtime: it can speak, sing, and even understand impromptu changes in wording.

Beating AI Flash News: Tencent Hyunyuan has released a brand-new real-time voice model, Hy Realtime, adding singing capabilities and online search during voice conversations. It has already been integrated into Yuanbao voice calls and Tencent Customer Service.

Users can directly use voice to have the AI search for the latest information, or request songs by singer, genre, language, and release year. For example, asking it to sing an English folk song from 2010 to 2020 allows the model to select and perform a song based on context. The singing feature was jointly developed by Hyunyuan and Tencent Music TME's Lingyin AI team.

The new model also specifically addresses the issue of voice AIs easily "losing intelligence upon speaking." Knowledge and reasoning abilities originally mastered by text-based large models may be lost when converted into voice conversation models. Tencent redesigned the training data production process, claiming that the effective migration retention rate of high-quality text data to voice tasks has increased from 21% to 69%, about 3.3 times the previous level.

Hy Realtime has also enhanced its understanding of natural spoken language. Even if users speak vaguely, only say half a sentence, or change their minds mid-conversation, the model can restore their intent by combining context. When generating responses, it can organize text while synthesizing speech simultaneously, reducing waiting time.

Currently, users can choose the "Chuyu" voice tone to experience this in the latest version of the Yuanbao App's voice calls. Tencent Cloud API plans to open soon, with duplex interaction and long-term personalized memory remaining as future upgrade directions.

Original article link https://m.theblockbeats.info/flash/377193

Key points

Evidence indicates that Tencent Hunyuan released the Hy Realtime real-time speech model.
The model has added online search capabilities for singing and voice dialogue, and has been integrated with Yuanbao voice calls and Tencent customer service.
Users can use voice commands to have the AI search for the latest information, and they can also request songs by artist, genre, language, and release year.
Evidence states that the singing feature was jointly developed by Hunyuan and Tencent Music TME’s Lingyin AI team.
Tencent says redesigning its training-data production pipeline raised effective retention when transferring high-quality text data to speech tasks from 21% to 69%, about 3.3 times the previous rate.
The source says the model uses context to understand ambiguous, unfinished or revised speech and synthesizes audio while composing its response.
Users can try the Chuyu voice in calls in the latest Yuanbao App. The Tencent Cloud API is planned for release soon.
Duplex interaction and long-term personalized memory are planned upgrades.

AI-assisted interpretation

The following is analysis, separate from reported facts. Verify important claims independently.

Hy Realtime is a real-time voice model. According to the material, it can search by voice, sing songs matching requests and respond to incomplete or revised speech. Users can try the Chuyu voice in the latest Yuanbao App.

Why it matters to readers

The model brings voice search, singing and voice conversations to Yuanbao calls and Tencent customer service. The Tencent Cloud API is planned for release soon, with no specific date or pricing given.

Users can select the Chuyu voice in calls in the latest Yuanbao App. The material gives no release date or pricing for the Tencent Cloud API.

Risks and unknowns

Tencent reported the 21% to 69% data-transfer retention figures without providing independent verification.
No official opening date or pricing is given for the Tencent Cloud API.
The material provides no usage adoption or commercialization results.
Duplex interaction and long-term personalized memory remain planned upgrades, and their progress is unspecified.

Tencent Hunyuan has released the Hy Realtime real-time speech model, which has been integrated into Yuanbao and Tencent Customer Service.

Article

AI-assisted interpretation

Why it matters to readers

Risks and unknowns

Related Developments

Related concepts

Keep following

Security Center