Chinese · 中文
Chinese AI voice generator and Mandarin voice cloning
- Language
- Chinese (中文)
- Script
- Han characters (Simplified and Traditional)
- Reference clip limit
- 12 seconds
- Reach
- over a billion speakers of Mandarin
- Available on
- Starter and above — the free tier generates English only
Two properties of written Chinese shape how you should type for it. First, tone is not written in Han characters at all — it is a property of the syllable the character represents, and the model has to know the character to know the tone. Second, some characters have more than one reading: 行 is xíng or háng depending on the word it sits in, and 长 is cháng or zhǎng.
Both are resolved from context, which is why full sentences generate more reliably than isolated words or fragments. If a single character comes out with the wrong reading, the fix is almost always to give it more sentence around it.
What is different about generating Chinese
- Reference recordings for this language are capped at 12 seconds, the same limit used throughout Mayantar.
- Both Simplified and Traditional characters are read. The script you type does not change the spoken output — Mandarin is Mandarin either way.
- Polyphonic characters (多音字) are disambiguated from context. Isolated words are the weak case; put the character in a full sentence.
- Pinyin typed in Latin letters is read as Latin text, not as Chinese. Type characters.
- Chinese does not use spaces between words, and none should be added — word segmentation is the model's job and inserted spaces disrupt it.
- Cloning is cross-lingual: a voice cloned from a clip in one language can generate in another on this path. It works, but it is not free — accent from the reference language carries over. Record in the language you intend to publish in unless a foreign accent is the point.
Recording a Chinese reference clip
Mayantar ships written prompts for English and the Indian languages; for Chinese the recorder offers the English prompts as a starting point, and you are better off reading Chinese of your own. Anything works as long as it is you talking normally — introduce yourself, describe what you are working on, read a paragraph of your own script.
Aim to fill most of the 12-second limit, keep it in one take, and include a question if you can: rising intonation is the thing a clone most often lacks when the reference was a flat, careful read.
The checks your clip has to pass — and the specific reasons clips get rejected — are in our guide to recording a reference clip.
Chinese voice generation: common questions
Does it support Cantonese?
No. The Chinese option generates Mandarin. Traditional characters are read, but they are read in Mandarin.
How does it know the tones if they are not written?
Tone is carried by the character, and the model learned character-to-sound mapping from speech. Ambiguous characters are resolved from surrounding context, so complete sentences are more reliable than fragments.
Is the Mayantar interface available in Chinese?
The interface and the generation language are separate settings. Mayantar's interface ships in eleven languages, and the audio it generates can be in any of thirty-four — so you can run the app in English and generate Chinese, or the other way round where both are offered.
Keep reading
- All 34 supported languagesThe full generation-language list with script requirements.
- How voice cloning worksFrom reference clip to a reusable voice, step by step.
- Consent and safetyWho may be cloned, what is watermarked, what we delete.
- Plans and limitsMinutes, characters and voice slots on each tier.
Generate Chinese in your own voice
Record a 12-second clip, type your script, download the WAV. Chinese generation is on Starter and above, from ₹299 a month.