option
Home
Flash News
Content
CarlPerez
CarlPerez
August 4, 2026

Tencent Hunyuan released Hy ASR3.0 preview, a speech recognition model powered by the Hy3 LLM. It improves general recognition, context awareness, multi-scenario robustness, and dialect coverage, evolving from word-by-word transcription to context understanding. On overseas open-source evaluation sets, word error rates reached 3.34% for Mandarin, 2.62% for English, and 3.12% for Cantonese. Tencent's own evaluation set also showed low error rates across various scenarios.

Tencent Hunyuan released Hy ASR3.0 preview, a speech recognition model powered by the Hy3 LLM. It improves general recognition, context awareness, multi-scenario robustness, and dialect coverage, evolving from word-by-word transcription to context understanding. On overseas open-source evaluation sets, word error rates reached 3.34% for Mandarin, 2.62% for English, and 3.12% for Cantonese. Tencent's own evaluation set also showed low error rates across various scenarios. Tencent Hunyuan released Hy ASR3.0 preview, a speech recognition model powered by the Hy3 LLM. It improves general recognition, context awareness, multi-scenario robustness, and dialect coverage, evolving from word-by-word transcription to context understanding. On overseas open-source evaluation sets, word error rates reached 3.34% for Mandarin, 2.62% for English, and 3.12% for Cantonese. Tencent's own evaluation set also showed low error rates across various scenarios.
Comments (0)
0/300
OR