中文(ZH) 腾讯混元Hy ASR 3.0 preview：让语音识别理解上下文

腾讯混元发布Hy ASR 3.0预览版，增强上下文理解能力

作者 PulseAugur 编辑部 · [3 个来源] · 2026-08-04 08:42

腾讯混元发布了其新的Hy ASR 3.0预览版，这是一款先进的语音识别模型，集深度语义理解与高精度语音识别于一体。该新模型在通用识别、上下文感知、多场景鲁棒性以及方言覆盖方面均有显著提升，在各项基准测试中实现了较低的词错误率。Hy ASR 3.0预览版现已通过API在腾讯云上线，并正被集成到元宝和WorkBuddy等产品中。 AI

影响通过提高语音识别的准确性和上下文理解能力，增强AI应用，有望带来更自然、更高效的人机交互。

排序理由大型AI实验室（腾讯混元）发布新模型。[lever_c_demoted from frontier_release: ic=2 ai=1.0]

在量子位 (QbitAI) 阅读 →

AI 生成摘要 · Google Gemini · 来自 3 个来源。我们如何撰写摘要 →

报道来源 [3]

量子位 (QbitAI) TIER_1 中文(ZH) · 允中 · 2026-08-04 08:58

Tencent Hunyuan Hy ASR 3.0 preview: Making speech recognition understand context

元宝已接入
36氪 (36Kr) TIER_1 中文(ZH) · 2026-08-04 08:42

Tencent Hunyuan releases latest speech recognition model Hy ASR 3.0 preview

36氪获悉，腾讯混元正式发布新一代语音识别模型Hy ASR 3.0 preview。在开源评测集中，Hy ASR 3.0 preview在多语种的WER（Word Error Rate, 词错误率）上都控制在3%左右，其中中文普通话WER 3.34%、英语WER 2.62%、粤语WER 3.12%。
Mastodon — fosstodon.org TIER_1 한국어(KO) · [email protected] · 2026-08-04 11:53

Tencent Hy (@TencentHunyuan) Hy ASR 3.0 preview has been released. Based on Hy3's language model, it enhances not only speech recognition accuracy but also context and meaning understanding, highlighting core performance in processing audio from various real-world environments and dialects, and robustness in multiple usage scenarios. http

Tencent Hy (@TencentHunyuan) Hy ASR 3.0 프리뷰가 공개됐다. Hy3의 언어 모델을 기반으로 음성 인식 정확도뿐 아니라 문맥·의미 이해를 강화했으며, 다양한 실제 환경의 오디오와 방언 처리, 여러 사용 시나리오에서의 견고성을 핵심 성능으로 내세운다. https:// x.com/TencentHunyuan/status/20 84579829303615497 # asr # speechrecognition # ai # languagemodel

报道来源 [3]

Tencent Hunyuan Hy ASR 3.0 preview: Making speech recognition understand context

Tencent Hunyuan releases latest speech recognition model Hy ASR 3.0 preview

相关实体

相关话题