Thai AI voiceover

The Thai voices, with samples

Each sample reads the same line so you can compare them fairly.

A segmenter where spaces do not exist

Most caption systems treat Thai as an unbreakable run and dump whole sentences at once. Here the pipeline runs each sentence through a dictionary word segmenter, the same segmentation the voice timing marks use, so the caption knows where words begin even though the script never shows it. The result: Thai captions type on word by word like English ones, and line wrapping lands between words instead of through them. It is worth one render to see, because it is the behaviour most Thai users assume they cannot have.

Eight th-TH voices and their lanes

Thai carries the complete premium cast. Puck and Laomedeia take promos and social spots at speed; Charon and Erinome deliver instructions patiently, with tutorials locked to them by the standing rule; Umbriel and Sulafat give the warm read a tea house or guesthouse wants; Iapetus and Aoede stay level for corporate work. The director casts style and gender from your brief and writes polite-particle register to match, khrap and kha landing where your described tone puts them. Scene-level swaps and full catalogue search live in each scene's voice dialog.

What the brief controls

Thai in the Language field is the entire mechanism: the director then writes narration and on-screen copy in Thai natively, from a brief in any language, and the th-TH locale stamps every automatic voice. Register is directable, formal for a clinic, playful for a street food stand. Thai phrases quoted in the brief arrive in the video exactly as typed. A language typed outside the mapped list would produce no voiceover at all, so use Thai as written in the suggestions rather than a variant spelling.

Timing, money and the fallback font

Scenes stretch to their clips and dynamic pace carries long lines over the next cut; per scene you have delay in seconds, speed from 0.85x to 1.25x, and the text box. Music ducks to 22 percent under speech on its own. Each synthesised clip is billed by length, and an export adds a small render charge once clips exist. The one visual caveat repeats: no look font ships Thai, so captions and Thai headlines take the fallback font's style. Keeping on-screen Thai brief, and letting the voice carry detail, plays to the system's strengths.

Captions in this script

Thai writes without spaces, and the captions deal with it properly: a dictionary-based segmenter breaks each sentence into real Thai words, joined without gaps in the pill, and the word-by-word highlight still tracks the voice through them. Long unspaced sentences are wrapped on those word boundaries at around 22 characters a line rather than mid-word. The glyphs come from the render host's fallback font, since no catalogue family ships Thai; above and below marks render correctly but in a style separate from the look's Latin. Short sentences keep both the segmenter and the pill at their best, so render once and read it back.

Questions people ask

Related pages