一名开发者在Hacker News上分享了新开源项目Yap，这是一款专为macOS设计的菜单栏应用，实现了无需下载模型的本地设备语音转文字功能[1]。该应用利用macOS 26新增的SpeechAnalyzer和SpeechTranscriber API进行语音识别[1]，用户只需设置快捷键便可进行语音输入，转录的文本随后直接粘贴到当前输入框[1]。
Yap采用MIT开源许可发布，代码完全用原生Swift编写，总计约3,000行[1]。应用完全在本地运行，不含任何网络代码[1]。在性能方面，Apple的语音模型在精度上略领先Whisper Small，同时处理速度快3倍[1]。从资源占用来看，应用体积仅为4 MB，空闲状态下的内存占用约60 MB[1]。
A developer has released Yap, an open-source macOS menu bar application that enables voice-to-text conversion entirely on users' devices without requiring model downloads [1]. The tool leverages the SpeechAnalyzer and SpeechTranscriber APIs introduced in macOS 26 to perform speech recognition locally [1]. After users configure a keyboard shortcut, they can dictate directly into the active input field, with transcribed text pasted automatically [1].
The application runs completely on-device with no network connectivity, ensuring privacy by design [1]. Written in native Swift comprising approximately 3,000 lines of code, Yap maintains a minimal footprint at 4 MB in size and consumes roughly 60 MB of idle memory [1]. According to the developer, Apple's speech recognition model achieves slightly better accuracy than Whisper Small while operating three times faster [1]. The project is distributed under the MIT open-source license [1].