Read aloud (TTS)
Listen to AI replies and page selections with three text-to-speech engines — the browser's built-in voices, local Kokoro 82M (WebGPU), or Volcengine cloud TTS.
Cordy can read text out loud — assistant replies in the side panel, or any selection on a web page. You choose the voice engine: the browser's built-in speech (zero setup), a local neural model that runs on your GPU, or a cloud voice.
Overview
Read-aloud is available in two places:
- On assistant messages in the chat — a Read Aloud control on the message bubble.
- On page selections — the Read aloud action in the floating menu, which opens a small player pinned to the edge of the page.
Turning it on
Enable Read Aloud (TTS) in settings, where you can also set the default Speed and pick the Read Aloud Engine. The engines and their model downloads live under Settings → Local AI → TTS Models (Browser TTS and Kokoro) and Cloud TTS (Volcengine).
The engines
| Engine | Setup | Requirement | Notes |
|---|---|---|---|
| Browser TTS | None | Always available | Uses your OS voices; quality varies by system |
| Kokoro 82M | ~330 MB download | WebGPU + Local AI enabled | Neural voices, on-device, 8 languages |
| Cloud TTS (Volcengine) | API key | Volcengine account | Large-model cloud voices |
Browser TTS (Web Speech)
The built-in fallback. It uses your operating system's speech synthesis, so there's zero download and it's always available. Voice quality depends on the OS and installed voice packs, and voices are chosen by language.
Kokoro 82M (local, WebGPU)
A neural text-to-speech model that runs entirely on your device through WebGPU, in an offscreen document. It's about a 330 MB one-time download and supports 8 languages (US and UK English, Spanish, French, Italian, Japanese, Brazilian Portuguese, and Chinese), each with a choice of voices. Requires WebGPU and Local AI enabled — see Local & on-device AI. If a language isn't covered (or WebGPU is unavailable), Cordy falls back to Browser TTS.
Cloud TTS (Volcengine)
Volcengine Speech Synthesis 2.0 (large-model TTS). Add your Volcengine API key (a single X-Api-Key — the latest scheme needs no App ID) in Cloud TTS (Volcengine) settings, choose a voice, and use Preview to test it. Note that each voice must be authorized in the Volcengine console before use. As a cloud engine, it sends the text to be spoken to Volcengine.
How Cordy picks an engine
Cordy uses the engine you select. In practice the resolution is: Cloud TTS if it's configured and you've consented, otherwise Kokoro if Local AI is on and Kokoro is your chosen local engine, otherwise Browser TTS as the always-available fallback. So even with nothing set up, read-aloud still works via your browser's voices.
Reading chat replies
On any assistant message, click Read Aloud. The control walks through loading → playing → paused, and you can pause, resume, or stop at any time — cancellation works even while a cloud or local model is still loading.
Reading a page selection
Select text, open the floating menu, and click Read aloud. A capsule player pins to the right edge of the page with Play / Pause / Replay / Stop and a progress bar. It stays put as you keep reading.
Controls, speed, and voices
- Controls: Read Aloud (play), Pause, Resume, Stop, and Replay.
- Speed: a single setting from 0.5× to 5× (default 1×), applied to whichever engine is active.
- Voices: Kokoro offers multiple voices per language; Volcengine has a cloud voice picker with a Preview button; Browser TTS uses the OS voice that matches the language.
Privacy
- Browser TTS and Kokoro run entirely on your device — no audio or text leaves it.
- Cloud TTS (Volcengine) sends the text to be spoken to Volcengine, and the first cloud use is covered by the one-time consent dialog.
Troubleshooting
- No sound. Check that Read Aloud is enabled and your system volume is up; try Browser TTS to rule out a model issue.
- Kokoro won't load. WebGPU is likely unavailable or Local AI is off — see Local & on-device AI. Cordy falls back to Browser TTS.
- Volcengine preview fails. Re-check the API key, and make sure the voice is authorized in the Volcengine console.
- Wrong language voice. Browser TTS depends on installed OS voices; install the voice pack for that language, or switch to Kokoro.