Sasayaku (囁く)

macOS menu bar app for push-to-talk speech transcription. Hold a key, speak, release — transcribed text is pasted wherever your cursor is. Fully local transcription powered by Whisper, optimized for Spanish.

How it works

Hold ⌥ Option → Record audio → Release → Whisper transcribes → Text pasted at cursor

Requirements

• macOS 14+
• Apple Silicon (M1/M2/M3/M4)
• Xcode Command Line Tools (xcode-select --install)

Install

make install    # builds release + installs to /Applications

Or build and run without installing:

make run        # builds release .app bundle and opens it

Other commands:

make build      # just build the .app bundle
make uninstall  # remove from /Applications
make clean      # clean build artifacts

Getting started

1. Launch the app

After make install, open Sasayaku from Applications or Spotlight. A mic icon (🎤) will appear in your menu bar.

2. Grant permissions

• Microphone: the system will prompt you automatically on first launch
• Accessibility: go to System Settings → Privacy & Security → Accessibility → enable Sasayaku. This is required for the global hotkey and text pasting to work.

3. Download a model

Click the mic icon in the menu bar. You'll see a yellow "Setup needed" status with a Download Model button. Click it and wait for the download to complete.

The default model is Medium (~1.5 GB) — it provides the best transcription quality for Spanish. If you have less than 16 GB of RAM, switch to Small (~466 MB) in Settings before downloading.

4. Start using it

Once the status turns green ("Ready"), hold ⌥ Option, speak, and release. The transcribed text will be pasted wherever your cursor is — works in any app.

Usage

Action	What happens
Hold ⌥ Option	Starts recording (icon turns green)
Release ⌥ Option	Stops recording, transcribes, pastes text (icon turns orange, then back to normal)
Click menu bar icon	Opens the popup with status, last transcription, and settings

Menu bar icon colors

• ⚪ Normal — idle, ready to record
• 🟢 Green — recording in progress
• 🟠 Orange — transcribing audio
• 🔴 Red — error (check the popup for details)

Models

Model	Download	RAM usage	Quality	Speed
Tiny	~75 MB	~200 MB	Basic	Fastest
Base	~142 MB	~300 MB	Good	Fast
Small	~466 MB	~850 MB	Great	Fast
Medium	~1.5 GB	~3 GB	Best	Default

Models are downloaded from HuggingFace on first use and stored in:

~/Library/Application Support/Sasayaku/Models/

You can change the model in Settings at any time. The new model will be downloaded if not already present.

Permissions

Permission	Why it's needed	How to grant
Microphone	Capture audio while recording	Automatic prompt on first launch
Accessibility	Global hotkey + paste text via Cmd+V	System Settings → Privacy & Security → Accessibility

Architecture

• UI: SwiftUI MenuBarExtra — menu bar only, no dock icon (LSUIElement)
• Transcription: SwiftWhisper (whisper.cpp via SPM)
• Audio: AVAudioEngine → 16kHz mono Float32 with automatic resampling
• Hotkey: NSEvent global/local monitor on flagsChanged
• Text insertion: Pasteboard + simulated Cmd+V via CGEvent (original clipboard is restored)

Project structure

Sources/Sasayaku/
├── App/           SasayakuApp, AppDelegate, AppCoordinator
├── Views/         MenuBarView, SettingsView, StatusIndicator
├── Services/      AudioCapture, Transcription, Hotkey, TextInsertion
├── Models/        AppState, WhisperModel
└── Utilities/     ModelDownloader, Permissions