Aria
Bright shonen-anime lead who narrates her own fight scenes
このページはまだ日本語になっていません。英語版を表示しています。
zh-CN · 简体中文
Chinese AI Voices
All forty characters perform in Mandarin, behind twenty-three distinct voices — the same count as French. Within one project no two characters are handed the same voice until all twenty-three are spoken for, so a Mandarin ensemble separates on its own. The work in Chinese is elsewhere: making sure every name is read with the reading you actually meant.
40人の登場人物のうち12人を、Chinese (Simplified)で
Koji
Rival character: clipped, competitive, secretly loyal
Lumen
VTuber persona: cute on top, deadpan underneath
Ezra
Visual-novel love interest with a two-second pause before honesty
Seraphine
Court princess whose politeness is a weapon
Kestrel
Low, weary swordswoman who has already lost this war once
Dorian
Audiobook baritone for literary fiction and slow reveals
Ashford
Velvet-voiced villain who apologises before he ruins you
Delphi
Oracle who answers the question you should have asked
Vex
Demon who negotiates in a reasonable tone
Gran Thea
Grandmother telling a folk tale she half believes
Orrin
Old man on a porch who remembers the flood
Twenty-three voices, and the order they arrive in
Twenty-three distinct voices puts Mandarin level with French and comfortably past what an ordinary scene needs. Casting inside a project will not hand two characters the same voice until all twenty-three are spoken for, so an ensemble separates itself and you do not have to police it — the constraint only bites on Plus and Pro, since Free caps a project at two cast members and Basic at five.
They are not all reached the same way, though, and that is worth knowing before you plan a large cast. Eleven of the twenty-three are natively Chinese voices, including Taiwanese Mandarin and regionally accented voices from Shaanxi and Liaoning, and the other twelve are multilingual voices that speak Mandarin without it being their first language. Every character's first choice comes from a hand-picked set of seven Mandarin voices, decided by its gender and age band: the senior women start on the Liaoning-accented voice, Milo and Pip on a real child's voice, and every young woman in the library on one shared voice. So no character starts on a borrowed voice in Mandarin — but seven is few, and a second character of the same gender and age band clashes the moment it joins. The one added later moves to the free voice closest to its age band, among its own gender first and then the other, and only a child character is ever moved onto the child voice; that is how the rest of the twenty-three are reached, and a second young-adult woman, for instance, lands on a Taiwanese Mandarin voice, while a second teenage girl lands on the Liaoning-accented one. Regional colour therefore follows a character's age, gender and clashes rather than anything you can ask for directly, and every character that has been moved is worth auditioning rather than trusting.
多音字, and respelling with a homophone
Polyphonic characters are ordinary in Chinese and concentrated exactly where fiction needs precision: in names. 单 is shàn as a surname and dān otherwise; 解 is xiè as a surname and jiě otherwise; 仇, 覃, 区 and 乐 behave the same way. Outside names, 行 and 重 and 长 change reading with sense, and a wrong pick is a wrong word rather than an odd inflection.
There is no reading-override list, and Chinese has no phonetic spelling equivalent to Japanese kana. What it does have is homophones: replace the misread character with one that has the same syllable, the same tone and only one reading, and the listener hears the name you meant — write 善 for the surname 单 and it comes out shàn. The cost is in the subtitles: the SRT export on Plus and Pro follows your script text, so the stand-in character appears there and has to be changed back in your editor for a subtitled release. Audition every name before you render an episode, because that is where the problem shows.
Tone sandhi breaks at the seams you create
Third-tone sandhi, 一 shifting tone with what follows, 不 before a fourth tone — the engine handles these inside a phrase and it handles them well. Where it fails is at boundaries you introduce: a line break or a comma dropped into the middle of a word stops the sandhi from applying, and 你好 read as two separate syllables is immediately audible to a native listener even though nothing is misspelled.
So the practical rule for Mandarin scripts is the opposite of the advice for most languages: do not split lines for pacing. Keep each utterance whole and let the line-level speed control on Plus and Pro handle the pacing instead, because that changes delivery without changing where the engine thinks the words end.
1シーンをキャスティングする手順
- Start the project in Chinese
A project takes its language from the CastDub interface you create it in and keeps it. Getting this right is also what stops a Japanese or Korean script being read with Mandarin readings of the shared characters.
- Add every speaker before you audition
Casting avoids repeats across the whole project, so which voice a character gets depends on who is already in the cast. Build the full cast list first and you will not have to re-listen to anyone.
- Audition the names
Every surname and every invented name, before you render anything long. Where the reading will not come right, swap in a character with the same sound and tone.
- Keep utterances whole
Do not break a line mid-phrase for rhythm. Use line-level speed control on Plus and Pro instead.
- Export
Mixed file on any plan, SRT from Plus, one track per character on Pro.
よくあるご質問
How many different voices can one Mandarin scene have?
Twenty-three. All forty characters perform in Mandarin, and casting inside a project avoids repeats until those twenty-three are used up, so twenty-three is how many speakers can sound genuinely different at once.
If I add a character later, do the others change voice?
No. Casting is first come, first served in cast order, so anyone already in the project keeps their voice and it is the newcomer who moves if there is a clash. That is deliberate — reshuffling would silently invalidate audio you had already rendered.
Is Cantonese supported?
No. Cantonese is a different spoken language rather than an accent setting, and written Cantonese is a different orthography as well. Pasting Cantonese text into a Mandarin project will not produce Cantonese.
Can I fix a wrong 多音字 reading?
Not with an override, because there is no reading list. Audition names early, and where one is misread, swap in a character with the same syllable, the same tone and a single reading — the listener hears the right name, though the SRT subtitles show the character you swapped in.
今夜、最初の1シーンをキャスティング
台本を貼り付ければ、CastDubが登場人物一人ひとりに声を割り当てます。あとは完成したボイスドラマを書き出すだけです。
無料で作ってみるクレジットカードは不要です。無料枠は一度きりの持ち分です。