YOUR AI PLATE HANDBOOK

Local voice: Moonshine & Kokoro

Set up speech input and spoken responses with local model workers.

Desktop source reviewed October 5, 2026

On this page

Install the voice assets

Open Plugins & Tools and enable the Moonshine STT and Kokoro TTS plugins you want to use.
Open each plugin’s configuration and download its model assets. The first download requires a network connection.
Wait until the model status is ready. The packaged Windows app includes Python; development checkouts may need npm run setup:python.
For Kokoro, choose a voice and speed, then use Test Voice.

Model downloads are separate from enabling a plugin. Use the displayed status and download progress instead of assuming the model is installed.

Choose the chat mode and model you intend to use.
Use the microphone button in the composer to record speech and check the transcription.
Use a response’s speech control to hear it with Kokoro, or enable Auto-Read Responses.
Adjust the voice and playback speed in the plugin configuration when needed.

LaTeX-to-speech processing prepares mathematical expressions for spoken output. Verify important formulas visually as well as by listening.

Symptom	Check
No transcription	Microphone access, enabled Moonshine plugin, and model/worker status.
No spoken response	Enabled Kokoro plugin, downloaded model assets, output device, and Test Voice.
Download interrupted	The download status and connection, then retry from the model controls.
Python worker unavailable	The resolved Python runtime; for source builds, run setup:python before testing again.

Implementation sources

Paths in your AI Plate source checkout: