Text to Speech
Uses the voices already installed on your device through the Web Speech API. Nothing is uploaded and no audio is generated on a server.
Worked example
The voice list is empty on first load in some browsers until the speech engine initialises. That is a platform quirk, not a failure — the list populates a moment later.
Speech synthesis has shipped in browsers for years and is still uneven: voice availability depends on the operating system, and the same page sounds different on Windows, macOS and Android.
What the voices are
These are the system voices, not cloud voices. Quality varies from the robotic default on a bare Linux install to neural voices on recent versions of macOS and Windows. Nothing is sent anywhere — which is the reason to use this over a cloud text-to-speech service when the text is private.
Practical settings
A rate of 1.0 is normal speed. 0.8 is useful for proofreading your own writing, because it forces you to hear the words rather than skim them. Pitch changes are mostly cosmetic; extreme values sound unnatural on every engine.
Limits
Long text can be cut off, because some engines cap utterance length. The workaround is to split into paragraphs and queue them, which is what production implementations do. This page reads what is in the box in one pass.