为什么值得关注

它利用 macOS 26 新增的 SpeechAnalyzer 和 SpeechTranscriber API，实现了比 Whisper Small 更低延迟和更高准确率的实时设备端转录；整个应用仅 4 MB、约 60 MB 空闲内存、纯 Swift 原生开发，不联网、无需账户或 API 密钥，音频完全留在本地。

适合谁使用

需要免提输入代码注释、提交信息或终端命令的开发者
希望快速听写邮件、Slack 消息等的 macOS 重度用户
注重隐私、要求离线设备端转录的用户
使用 Intel Mac 或老旧硬件、难以运行 Whisper 等重型模型的用户

典型使用场景

在编写代码或 prompt 时用语音输入，解放双手
比打字更快速地撰写邮件和 Slack 消息
在不切换应用或使用复杂界面的情况下听写笔记或文档
为打字困难或偏好语音输入的用户提供辅助功能

项目优势

轻量快速：应用仅 4 MB，空闲内存约 60 MB，无浏览器引擎，Swift 代码约三千行
设备端隐私：无网络请求、无账户、无遥测，音频从不离开设备
零模型下载：直接使用 macOS 26 自带的 SpeechAnalyzer 和 SpeechTranscriber，根据引用的基准测试比 Whisper Small 更快更准
全局快捷键可在任意文本输入字段工作；采用剪贴板恢复机制正确应对 Chromium 系应用的异步读取问题

使用前须知

需要 macOS 26 (Tahoe) 或更高版本，不兼容旧版 macOS
未沙盒化（需要辅助功能权限进行粘贴），且通过 Mac App Store 外分发
语言跟随系统区域设置，暂未提供内置语言选择器

README 快速开始

Yap

Blazing-fast voice dictation for macOS that works anywhere you can type.

Press a shortcut, talk, press it again. Your words land in whatever text field you were using. No account, no API key, no audio leaving your machine.

Built by Frigade.

https://github.com/user-attachments/assets/9bd8dae6-49d6-4c06-98b9-878e24dd401f

What it does

Yap lives in your menu bar and waits for a shortcut. Trigger it and a small window appears near the bottom of the screen with a live waveform and a running preview of what you have said so far. Press the shortcut again and the text gets pasted into the app you were already working in. Every transcript is saved locally, so you can go back and copy something again later.

The default shortcut is ⌘⇧D. You can rebind it, or set a single modifier key instead. Tapping right shift on its own works nicely if you have a spare thumb.

Install

Homebrew

brew install --cask frigadehq/tap/yap

To update later:

brew upgrade --cask yap

Direct download

Grab the latest .dmg from Releases and drag Yap into your Applications folder.

Released builds are signed and notarized, so macOS opens them without complaint.

Why we built this

There's no shortage of voice to text tools for macOS, and some of the open source ones are genuinely good. Most of them still run into some mix of the same problems:

Some cost money, which is a lot to ask for something that's now built into your OS for free.
Most make you download a heavy model. Whisper weights run to hundreds of megabytes, sit in your RAM, and only feel fast on a recent, high end Mac. It gets especially bad on Intel Macs, where models like Whisper and Parakeet often crash or refuse to run outright, and when they do run, a single paragraph can take thirty seconds to a minute to come back.
A lot are Electron or web-stack apps, so a whole browser engine idles in your memory just to run a menu bar icon and a settings window.
Plenty are bloated with settings and modes you will never open.
Some are closed source, so you are trusting that your audio and transcripts stay on your machine, with no way to check.

What changed recently is macOS 26. It added two APIs, SpeechAnalyzer and `S

项目描述

Free, open source voice dictation for macOS. On-device transcription with Apple's Speech framework. No cloud, no API keys, no account.

yap