The model from Building and Evaluating Fixed-Voice Thai TTS from Synthetic Speech — 82M parameters, twelve voices, not one of them ever recorded. It synthesises here on two vCPU, no GPU: the same as your PC, or slower. Type Thai, pick a voice, listen.
Warming up — the model loads in the background.
Thai is written without spaces between words, so the frontend segments it and converts to IPA before the model sees anything. If a line sounds wrong, read the phonemes first — that is usually where it went wrong.
Each voice ships with a line written to its persona and register, and verified to phonemize cleanly before release. Pick one to load it above.