Traditional Chinese AI voiceover

The Chinese (Traditional) voices, with samples

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

What splits traditional from simplified here

The two Chinese entries in the picker differ in locale, script and voice tier at once. Traditional resolves to cmn-TW: the director writes every line in traditional characters and the voices carry Taiwan Mandarin pronunciation, the natural pairing for Taiwan and a workable one for Hong Kong Mandarin audiences. The catch is the tier, since cmn-TW ships no premium characters and falls back to two Wavenet voices. Simplified, by contrast, gets eight premium voices on cmn-CN. Script loyalty or voice depth is the actual choice being made.

Working with a two-voice cast

Style requests collapse to gender here: energetic, warm, instructional and neutral all land on Wavenet-B for male or Wavenet-A for female. The register survives in the writing instead, and the director leans into it, promos arriving as short exclamatory bursts, tutorials as numbered even steps. Wavenet's flatter prosody actually suits instructional content decently; it is long emotive passages where the age shows. The per-scene voice dialog searches the full catalogue for cmn-TW's remaining options, and a scene could even borrow a cmn-CN premium voice if mainland pronunciation is acceptable for that line.

The pipeline around the voices

Everything nonvocal runs at full strength. Briefs in any language work, with the Language entry alone deciding script and locale. Scenes stretch to their clips, dynamic pace floats overruns across cuts, and each scene exposes a delay in seconds plus speed steps from 0.85x to 1.25x, the slower end worth trying on dense lines. Music ducks to 22 percent beneath speech automatically. Synthesis is billed by length, about half a credit for a typical clip; the export bills its render time, usually a small amount.

Characters as your pronunciation control

No phonetic input exists, and zhuyin is not accepted either; characters do the work. Transcribe foreign names into phonetic characters in the narration, the convention Chinese copy already uses, and the voice reads those syllables and tones, about half a credit per regenerated clip. Write numbers as characters when their phrasing matters, 兩百五十元 being explicit where 250元 is not, and note Taiwan's 兩 preference the director follows naturally. Sentence punctuation cuts caption cues, and short sentences suit the flatter Wavenet read as much as the pill.

Captions in this script

Traditional Chinese captions behave like simplified ones mechanically: sentences cut on 。！？ punctuation are segmented into dictionary words, joined without spaces in the pill, wrapped between words at roughly 22 characters, with the highlight walking the words in time. A Wavenet-specific bonus applies here: these voices honour timing marks directly, so word sync is measured on the exact clip you hear. The traditional glyphs come from the render host's fallback font, no catalogue family shipping Han characters, so expect the caption's typeface to sit apart from the look's Latin type.

Questions people ask

Related pages