Chinese AI voiceover

The Chinese (Simplified) voices, with samples

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

Two Chinese entries, two different tiers

The Language picker lists Chinese twice, and the split is not cosmetic. Simplified maps to cmn-CN with the complete premium cast of eight, the strongest Mandarin offering. Traditional maps to cmn-TW, where the catalogue holds no premium voices at all and two Wavenet voices stand in. So if your market reads simplified characters, this entry gives you both the modern script and the modern voices; the traditional entry, covered on its own page, trades voice quality for the script Taiwan and Hong Kong read. The director writes whichever script the entry names.

Casting Mandarin across four styles

Puck and Laomedeia bring livestream-commerce energy for promos and social spots. Charon and Erinome pace themselves for instruction, with tutorials permanently assigned to them by rule. Umbriel and Sulafat suit testimonial warmth and food storytelling. Iapetus and Aoede read corporate updates plainly. All eight speak standard Mandarin with mainland pronunciation, erhua where standard. Style plus gender is chosen from your brief; any scene re-casts through its voice dialog, and the search there spans the entire live Google catalogue, including other Chinese locales.

Names, tones and the character fix

There is no pinyin input, and none is needed: characters are the pronunciation control. A foreign brand read awkwardly gets a phonetic character transcription in the narration, the way Chinese marketing already writes 星巴克 for Starbucks, and the voice reads those syllables with their tones, about half a credit per regenerated clip. Homograph characters with two readings resolve the same way, choose a synonym or restructure the phrase. Numbers do best written as characters when phrasing matters, 一百五十元 rather than 150元, since digits let the engine pick.

Density, duration and what you pay

Mandarin is compact; a sentence often speaks faster than its English draft, which suits 15 second promos well. Where a line runs long the scene stretches to its clip, and dynamic pace plays the overrun across the next cut. Per scene: delay in seconds, speed steps from 0.85x to 1.25x, and the narration box. Music self-ducks to 22 percent beneath the voice. Synthesis is billed by length, about half a credit for a typical clip, and the export bills its render time, needing at least 1 credit left past 30 seconds or with lines still pending.

Captions in this script

Chinese runs without spaces, so the caption pipeline segments each sentence, cut on 。！？ punctuation, into dictionary words and joins them without gaps in the pill; the highlight then follows the voice through those words in order. Unspaced lines wrap on word boundaries at about 22 characters. The characters themselves are drawn by the render host's fallback font, because no catalogue family ships Han glyphs, so caption typography will not match the look's Latin. Short sentences keep each cue small, which flatters both the segmentation and the pill.

Questions people ask

Related pages