SelectSpeak: Read Aloud Text to Speech (TTS)

Read any selected text aloud with a natural AI voice. Private, on-device text to speech (TTS) — no text ever leaves your machine.

Select text on any page and hear it read aloud in a natural neural voice entirely on your device. No cloud, no accounts, no data collection. Your text never leaves your machine. SelectSpeak was built for readers who find long articles easier to follow by ear including readers with mild dyslexia. The paragraph currently being spoken is highlighted on the page and auto-scrolled into view, so your eyes and ears stay in sync. HOW IT WORKS 1. Select text on any page. 2. Click the Listen pill that appears (or press Cmd+Shift+Period / Ctrl+Shift+Period, or click the toolbar icon). 3. A mini player gives you play/pause, skip ±10s, a progress bar, a voice picker, and a 0.5x–4x speed slider. FEATURES • Natural on-device neural voice (Kokoro-82M) — 4 voices, US and UK, female and male • Live paragraph highlighting with auto-scroll • Clear speech even at 3x–4x: pauses between words and sentences are preserved, so fast playback stays articulate instead of smearing • Speed changes resume from the sentence you were on, not the top • Keyboard shortcut for hands-free start (cmd + shift + .) PRIVACY — THE WHOLE POINT • All text-to-speech runs locally on your machine (WebGPU, falling back to WASM) • No analytics, no telemetry, no accounts, nothing collected — ever • The only network request the extension can make is a one-time download of the open-source voice model from huggingface.co, verified against checksums pinned at build time and then cached locally. After that: zero network requests, enforced by an automated check in the build FIRST RUN The first time you use it, the voice model downloads (~90 MB, or ~330 MB for the higher-quality WebGPU version) and is cached. Later reads start instantly. OPEN SOURCE Code, privacy policy, and issue tracker: https://github.com/dave-song/selectspeak SelectSpeak uses the open-source Kokoro-82M model (Apache 2.0) via kokoro-js.