sayd

Local text-to-speech for Wayland. Select text, press a key, hear it — with nothing leaving your machine.

Kokoro-82M runs locally through ONNX Runtime with a misaki-derived English frontend. There is no cloud inference, no telemetry, and no network access at runtime.

Status

Early. The engine and audio path work: text in, chunk-streamed speech out, starting about a second after the keypress regardless of how much was selected. The D-Bus interface, control CLI, tray icon and settings window are in progress.

Requirements

• Wayland compositor supporting wlr-data-control v2 for the primary selection (sway 1.9+)
• ONNX Runtime at ORT_DYLIB_PATH
• espeak-ng, with ESPEAK_DATA_PATH set
• Kokoro model weights and voice packs in models/

Licence

MIT, except the vendored misaki lexicons in sayd-misaki-en, which are Apache-2.0.