Voice lab

Make the voice cute.

The default is your device's own voice, with pitch, speed, wobble and breaths turned up to cute. An optional local server adds the better voices (Qwen3-TTS, then Kokoro). Nothing plays until you press a button.

Controls

Try a line

Nothing spoken yet.

Your own line

The better-voice server

Checking http://127.0.0.1:4650 ...

Start it (Apple silicon Mac, one-time install in ~/.kamiji-voice, models are cached in ~/.cache/huggingface):

tools/voice-server/run.sh          # start (127.0.0.1:4650)
tools/voice-server/run.sh status
tools/voice-server/run.sh stop

Then choose “Better voice” in the controls. If the server is not running the page keeps using the system voice. Qwen takes a few seconds per sentence; Kokoro is faster.