Tencent Hunyuan released its next-generation speech recognition model Hy ASR 3.0 preview today.
According to the official announcement, leveraging the language understanding capabilities of the latest large language model Hy3, Hy ASR 3.0 preview combines high-precision speech recognition with deep semantic understanding. It delivers comprehensive improvements in key areas including general recognition, contextual awareness, robustness across multiple scenarios, and dialect coverage, providing accurate, coherent transcriptions that better reflect user intent for more complex real-world inputs. It evolves from “verbatim transcription and point-based optimization” to “understanding context, adapting to scenarios, and producing results with one click.”
Hy ASR 3.0 preview delivers leading overall performance across multiple open-source and in-house evaluation datasets. On open-source evaluation datasets, Hy ASR 3.0 preview keeps multilingual WER (Word Error Rate) at around 3%, including a Chinese Mandarin WER of 3.34%, an English WER of 2.62%, and a Cantonese WER of 3.12%.
On in-house evaluation datasets, Hy ASR 3.0 preview also maintains a low WER in scenario-based evaluations covering general recognition, dialect recognition, contextual understanding, specialized-term understanding, and complex acoustic environments such as high noise and whispering.
According to the official announcement, Yuanbao was deeply involved in the joint development and has completed its first launch. Users can open Yuanbao and press and hold to speak to experience improved dialect recognition, intelligent contextual correction, and stable transcription in complex environments. Voice input is now more accurate and stable, and is available for free. Products such as WorkBuddy are also being integrated gradually.
Hy ASR 3.0 preview is now available on the Tencent Cloud official website with API services, and can be widely applied to scenarios such as intelligent customer service, content understanding, and voice search.
Related links:
Tencent Hunyuan: https://aistudio.tencent.com/visual
Tencent Cloud: https://cloud.tencent.com/document/product/1093/135476