Catalog voices only · No cloning

Chinese text to speech

Written Chinese places no spaces between its words, so before a single syllable can be produced the text must first be divided into units. Divide it wrongly and the sentence is not merely mispronounced — it has been parsed into different words altogether, which is a far more destructive error than any accent slip could be.

Try the live product

Type, choose a voice, then press play.

Live demo

Type a script, pick a voice, and hear it read back.

Free to try — no sign-up.

44/500
Voice

Tap a voice to hear it instantly, then press play to hear your script in it.

A substantial number of characters carry more than one reading, selected by context and by the compound a character happens to sit inside. A surname and an everyday noun can share one glyph while sounding nothing alike. Settling those readings is the central task, and it cannot be done by looking at characters one at a time in isolation.

Tone completes the picture. Mandarin separates words by pitch contour, so a reading that is perfectly segmented yet tonally flat still fails the listener. Here tone is treated as an intrinsic part of each word rather than as expressive colour applied afterwards.

No signup needed — the demo speaks up to 500 characters.

Regional delivery

  • Chinese (Simplified)

Every voice speaks every supported language, so one narrator can carry your whole catalogue across markets.

Chinese (Simplified) production details

  • Segmented before voicing

    Unspaced text is divided into word units first, because an incorrect division changes which words end up being spoken at all.

  • Multi-reading glyphs settled

    Characters carrying several possible pronunciations are resolved from surrounding context instead of defaulting to the commonest one.

  • Tone as part of the word

    Pitch contours are reproduced as lexical information, since Mandarin uses them to separate vocabulary rather than to add expression.

  • Names fixed by rule

    Surnames and place names taking an irregular reading can be saved once and then applied to everything generated later.

56 studio voices (30 premium, paid plans) · 20 output languages · SRT captions and bulk mode on every plan

Free to start

Use the free plan first, then move to a monthly plan when your production volume grows.

25,000 characters

Free plan monthly allowance

$7/month

About 3 hours on the Creator plan

Compare all plans →

Questions, answered

Does it take simplified or traditional characters?

Simplified characters are the supported input here. Traditional text follows a different writing convention and is best converted before generation if that is what your source material uses.

How are characters with several readings resolved?

The reading follows the surrounding context and the compound the character appears within, rather than falling back on whichever pronunciation is most frequent when the glyph stands alone.

Why does word segmentation matter so much here?

Because the writing system carries no spaces, segmentation determines which words exist in a sentence at all. A wrong split does not create an accent problem, it creates different vocabulary.

Can I control how a surname is pronounced?

Yes. A number of common surnames take readings that differ from the same character's everyday pronunciation, so saving a rule for the names in your material is worth doing once up front.

Is tone genuinely reproduced?

Yes, and it must be. Tone distinguishes words in Mandarin, so a flat delivery would merge items of vocabulary that listeners hear as completely separate from one another.

Your next project

Publish in Chinese (Simplified) this week

Start free