Mockingbird

Select text anywhere on your Mac and hear it read aloud — natural neural voices, fully on-device.

Mockingbird is a small menu-bar app: select text in any app, press Ctrl+Option+S, and listen. Pause and resume with Ctrl+Option+P. Speech is generated locally by the voice engine of your choice — no accounts, no cloud, and after the one-time engine download it works fully offline.

Download for macOS 14+ (Apple Silicon) — signed and notarized, auto-updates via Sparkle. Or build it yourself below. MIT licensed.

What It Owns

• First-run onboarding that explains the app and downloads a voice engine of your choice.
• Global hotkeys, no Apple Shortcuts app required.
• Cold-start synthesis: no speech process stays resident while idle.
• Menu bar controls for engine choice, generation state, playback progress, pause/resume, stop, voice settings, audio cache, and editable hotkeys.

Default hotkeys:

• Read selection / stop current reading: Control + Option + S
• Pause/resume: Control + Option + P

You can edit them from the Mockingbird menu.

Voice Engines

The app ships tiny and nothing is self-hosted. On first launch, onboarding asks the user to pick an engine and builds it on the fly:

• Kokoro (recommended) — the most natural neural voices. Model weights and voices fetched from Hugging Face (hexgrad/Kokoro-82M). ≈1.3 GB installed.
• Piper (fast & light) — compact ONNX voices, quick on modest hardware. Voices fetched from rhasspy/piper-voices. ≈470 MB installed.
• Apple System Voices — no download; uses the voices built into macOS.

For Kokoro and Piper, scripts/engine_setup.sh bootstraps uv (a single static binary from astral.sh), creates a private Python 3.12 venv under the app's runtime folder, and installs version-pinned packages from PyPI (python/requirements-<engine>.txt). No system-wide installs, no Homebrew, no pre-existing Python required. Model weights are downloaded by the app with byte-level progress.

Engines live in ~/Library/Application Support/Mockingbird/engines, models in .../models. Users can switch, install, or uninstall engines any time from Settings > Voice Engine (menu bar icon > gear > Settings).

Build And Launch

scripts/package_app.sh
scripts/launch.sh

The packaged app lives at:

Mockingbird.app

Ship To Another Mac

Build a DMG, then share dist/Mockingbird.dmg:

scripts/export_dmg.sh

The recipient opens the DMG, drags Mockingbird.app to Applications, and launches it. Onboarding walks them through downloading a voice engine — they do not need Python, Homebrew, or developer tools installed.

You can also create a zip archive with:

scripts/export_app.sh

Local builds are ad-hoc signed by default. For a DMG intended for other Macs without Gatekeeper workarounds, sign with a Developer ID Application certificate:

MOCKINGBIRD_CODESIGN_IDENTITY="Developer ID Application: Your Name (TEAMID)" scripts/export_dmg.sh

Legacy packaging modes

MOCKINGBIRD_BUNDLE_SPEECH_HELPER=1 scripts/package_app.sh   # old self-contained app with Kokoro baked in (~900MB)
MOCKINGBIRD_REUSE_SPEECH_HELPER=1 MOCKINGBIRD_BUNDLE_SPEECH_HELPER=1 scripts/package_app.sh  # reuse an already-built helper
MOCKINGBIRD_SKIP_SPEECH_HELPER=1 scripts/package_app.sh     # dev build that installs a Python venv on first run

Local Runtime

Mockingbird keeps engines, models, generated audio, and request files in its Application Support folder, not in this repository. Local helper scripts such as scripts/read-selected.sh, scripts/pause.sh, and scripts/stop.sh send commands to that runtime folder.

Privacy

Mockingbird does not listen to your microphone. It only reads selected text when you press the read hotkey. Speech is generated locally; after the one-time engine download, no internet connection is used.