AMAZINGINDEX.COM 让生活多一点 Amazing
60.9
VOL. 2026.10
2026.10.09
← 返回 2026.10.09 日报
日报快照 · Daily Snapshot
NO. 013

16.9MB 离线语音模型跑在 CPU 上

#ARTICLE HackerNews 2026.10.09
推荐指数 73.0 NO. 013 · 2026.10.09
发布2026/10/08Score417Comments101

Whistle 发布端侧语音识别模型,单文件 16.9MB、零依赖、CPU 运行,支持 7 种语言转写、词级时间戳和语音嵌入。对需要离线语音能力的 IoT、车载、穿戴设备开发者,这是 Whisper.cpp 之外更轻量的替代方案。

端侧语音之前基本是 Whisper.cpp 的天下,但它 39MB 起步且需要 ffmpeg 等依赖,在资源极紧的 MCU 和穿戴设备上仍显笨重。Whistle 把体积压到 16.9MB 还做到零依赖,关键是和 Needle 共享同一套 C++ 推理引擎和量化方案,说明团队在做系统级的端侧 infra 而非单点模型。

做智能家居、车载语音助手的团队可以重点关注,尤其是需要唤醒后本地识别隐私敏感场景(如医疗穿戴)。语音嵌入输出也是个隐藏亮点,适合做声纹识别或语音搜索的下游任务,不需要额外跑一个 encoder。

意见分歧 152 条评论

核心争论:轻量端侧模型实用性与专用场景(如方言、老年语音、自然对话节奏)的识别质量差距

andy_ppp

Wow certainly in English this is incredibly accurate I tried to break it and it understood me perfectly! I know it's slightly off topic but surely it must be easy by now to train a spell checker that doesn't annoy the crap out of everyone using it (looking at you here Apple)!

tecleandor

Spanish is not good (seems to write non existing words and/or with terrible typos...) but English seem to work good even with my (Spanish) accent...

kaoD

Spanish from where? Here (Castilian Spanish) it seemed to work fine.

替代方案: Whisper.cppHandyParakeet streamingGemini 3.5 TranscribeFluidVoiceQwen3.5 9bWindows speech to textClaudeOpenAI
查看原文 →