Yap – Open‑source on‑device voice dictation for macOS (no model download)

Yap – Open‑source on‑device voice dictation for macOS (no model download)

Overview

Yap provides free, open‑source voice dictation for macOS that works entirely on‑device, requiring no model download, no API key, and no network traffic.

How Yap Works

Yap lives in the menu bar, activates with a configurable shortcut, captures audio via AVAudioEngine, and transcribes it using Apple’s SpeechAnalyzer API (with SFSpeechRecognizer as a fallback) to produce live, streaming text that is pasted into the currently focused field.

Installation and Setup

You can install Yap via Homebrew with brew install --cask frigadehq/tap/yap or by downloading the signed .dmg from the Releases page and dragging Yap into Applications.

Features

  • Global, rebindable shortcut (default ⌘⇧D) with optional single‑modifier triggers.
  • Live waveform and partial transcript while speaking.
  • Text is pasted directly into the active app using clipboard and simulated ⌘V.
  • Local transcript history with search, copy, and delete.
  • Follows the system default microphone, even if it changes mid‑session.
  • Optional launch at login, off by default.
  • No account, no network calls, no telemetry.

Comparison with Built‑in macOS Dictation

Unlike the built‑in dictation feature, Yap does not rely on a separate model download, runs entirely in a native Swift app (~4 MB), and provides a persistent transcript history and a menu‑bar interface.

Performance and Benchmarks

The README cites a benchmark that reports Apple’s on‑device Speech framework achieving a 2.12% word error rate on clean audio and 4.56% on noisy audio, compared to 3.74% and 7.95% for Whisper Small, and running about three times faster across 5,559 LibriSpeech clips.

Community Feedback and Known Issues

  • A user reported that the Homebrew tap URL does not exist: "remote: Repository not found. fatal: repository ' https://github.com/frigadehq/homebrew-tap/' not found"【__erik】.
  • Another user questioned how Yap differs from the built‑in macOS dictation feature【heresalexandria】.
  • A comment linked a benchmark comparing Apple’s Speech API to NVIDIA Parakeet TDT v2, noting Parakeet still leads but Apple is close【eric_khun】.
  • Users noted a similarly named project (finnvoor/yap) that also uses the new Speech APIs【hbn】.
  • Suggestions included removing the dock icon, adding a Settings… menu with Command-, shortcut, allowing F13 as a shortcut, and making transcript history ephemeral【jorisw】.
  • On macOS 27 Beta (build 26A5388g) a user experienced no transcription after speaking, a delayed waveform, and a crash【jorisw (second comment)】.
  • Some users expressed interest in adding custom voice commands for app launching or navigation【prabhanjana_c】.
  • One user mentioned having built a similar tool and seeing it as a potential future direction【solos】.
  • Several commenters expressed excitement about on‑device speech recognition for privacy and speed【echoscarrie】.

Roadmap and Future Work

The project states that the roadmap is intentionally minimal; the main design principle is keeping Yap small and focused. A language picker is mentioned as the most likely next addition, since Yap currently follows the system locale.

Conclusion

Yap offers a lightweight, privacy‑first dictation solution for macOS that leverages Apple’s newest on‑device speech APIs, eliminates model downloads, and provides a simple menu‑bar workflow, while community feedback highlights areas for improvement such as Homebrew installation, shortcut flexibility, and optional features like custom commands or ephemeral history.

Sources