Chinese Text to Speech: Natural Mandarin AI Voices
Chinese text to speech with 15 native Mandarin voices: paste Simplified or Traditional characters, get an MP3 with subtitles, and fix tones and polyphones.
Last updated: 2026-10-26
Chinese text to speech turns written Chinese into spoken Mandarin: paste Simplified or Traditional characters, pick a voice, and download an MP3 with subtitles timed to every character. VoiceSmiths has 15 native Mandarin voices made for Chinese content (14 original character voices plus Yun, a broadcast anchor from the ElevenLabs library) and renders them on Eleven v4. Every sample on this page is free to hear; generating your own script uses credits.
Hear the Mandarin voices
Tap a voice to hear it. To voice your own text, paste it into the workbench: 7 credits per 100 characters, starting with the $3.90 trial pack.
Hear Mandarin text to speech
An everyday paragraph, the kind a learner or a travel video needs, read by Composed Lady:
The same voices carry news, stories and fiction. News Anchor opening a newscast, then Old Storyteller in the pingshu style and Wandering Swordsman opening a fantasy chapter:
Two characters in one request, with audio tags that are acted and never read aloud:
[curious] 你这猴头，又从哪里闯了祸回来？
[laughs] 师父莫急，俺老孙不过是借了他三根毫毛！
[sighs] 借？只怕人家此刻正满山找你呢。
[excited] 找便找，俺一个筋斗，早到了十万八千里外！Mandarin voices by job
| Job | Voice | Why it fits |
|---|---|---|
| News, corporate, explainers | News Anchor, Yun | Standard broadcast Mandarin, crisp and even |
| Courses and training | Composed Lady | Low, warm, unhurried |
| History, folk tales, audiobooks | Old Storyteller | The rise and fall of traditional pingshu |
| Martial arts and fantasy fiction | Wandering Swordsman | Deep and weathered |
| Romance and short drama | Domineering CEO, Ice Queen, Gentle Healer | The three archetypes of Chinese short drama |
| Late-night, emotional, podcasts | Late-Night DJ | Close-mic, soft, long pauses |
| Ads and short videos | Gentle Healer, Livestream Host | Soft sell or hard sell |
| Kids' stories | Playful Girl, Big Bad Wolf | A hero and a villain for any bedtime story |
| Games, sports, hype | Hot-Blooded Youth, Esports Caster, Monkey King | High energy |
More pairings by use case are in best Chinese AI voices, and the formal readers are grouped under news anchor voices.
Which model reads Mandarin best
We sent the same Mandarin paragraph to Eleven v4 and Eleven v3 three times each, and once to Google's Gemini 3.8 TTS, then scored every take against the script with speech-to-text. Eleven v4 read it with no errors in all three runs. Eleven v3 heard 眼镜 (glasses) as 眼睛 (eyes) in two of three. Gemini made the same slip in its one take. Same paragraph, one take from each:
Eleven v4:
Eleven v3:
Gemini 3.8 Flash TTS (for comparison; the studio renders on ElevenLabs models):
The full numbers are on Eleven v4 vs v3. Pick Eleven v4 in the text-to-speech tab (the link at the end of this page opens it already selected); the audio drama studio renders every drama on it.
Tones and polyphonic characters
Mandarin has four tones plus a neutral one, and the model takes them from the characters and the context, not from anything you mark. Most errors come from polyphonic characters (多音字): one character, two or three readings, chosen by the word it sits in.
This test sentence uses 行 five times in two readings, plus 长, 得 and 重. Listen for 行长 hángzhǎng, 银行 yínháng, 一行人 yì xíng rén, 不行 bùxíng, 得 děi, 重新 chóngxīn and 一行 yì háng:
| Character | Readings | In context |
|---|---|---|
| 行 | háng / xíng | 银行 yínháng (bank) · 不行 bùxíng (won't do) |
| 长 | zhǎng / cháng | 行长 hángzhǎng (bank president) · 长城 Chángchéng (Great Wall) |
| 重 | chóng / zhòng | 重庆 Chóngqìng · 重要 zhòngyào (important) |
| 得 | de / dé / děi | 跑得快 pǎo de kuài · 得到 dédào · 得走了 děi zǒu le |
| 还 | hái / huán | 还有 háiyǒu (also) · 还书 huán shū (return a book) |
| 乐 | lè / yuè | 快乐 kuàilè (happy) · 音乐 yīnyuè (music) |
Names and places are the hardest, because nothing in the sentence tells the model they are special. As surnames, 单 is Shàn, 曾 is Zēng, 解 is Xiè, 仇 is Qiú and 区 is Ōu. Place names keep old readings: 六安 is Lù'ān, 蚌埠 Bèngbù, 铅山 Yánshān, 台州 Tāizhōu. Hear the wrong and right readings side by side for 六安, 蚌埠 and 重庆, and more in the pronunciation guide.
How to fix a misread character
- Respell it with a homophone. Swap the character for one that only has the reading you want: 重庆 → 虫庆, 六安 → 陆安, the surname 曾 → 增. No pinyin or phonetic alphabet needed.
- Put the respelling in a dictionary once. In the text-to-speech tab, add the rule to a pronunciation dictionary and tick it. Every later generation reads it your way, and the respelling never appears in the subtitles. Test it on one short line first.
- In long-form and full-cast drama, respell in the text. Dictionaries apply only in the text-to-speech tab (up to 5,000 characters per generation). For a long-form audiobook or an audio drama, type the homophone into the text itself; the subtitles will then show the homophone, so correct the SRT if you publish it.
Writing Chinese text that reads well
- Paste characters, not pinyin. The models read written Chinese. To hear a pinyin line, convert it to characters first (any pinyin keyboard does it) so the tones come from the words.
- Punctuation sets the pauses. ，and 。 give short and long pauses; …… and a line break give longer ones. Put a key sentence on its own line to make it land.
- Write numbers the way they should be spoken. 2026年 can come out as a year or a quantity; write 二〇二六年 if it matters. In phone numbers, people say 幺 yāo for 1, so write it if you want it.
- English words get a Chinese accent. Brand names and English terms inside a Chinese sentence are read by a Mandarin voice; respell them or add them to the dictionary.
- Audio tags work in Chinese text. Put
[whispers],[laughs]or[sighs]before the words they color; they are performed, not read, and never reach the subtitles. See the emotion tags cheat sheet. - For learners: the voice settings in the text-to-speech tab include a speed slider from 0.7× to 1.2×, and the SRT subtitles are built from per-character timestamps, so the text stays in step with the audio.
Accents: Mainland Mandarin, Taiwan and Cantonese
All 15 voices speak Mainland Mandarin. Traditional characters are read too, in the same accent, so a Taiwan-market script will sound Mainland, with Mainland readings where the two differ (垃圾, for example, is lèsè in Taiwan and lājī on the Mainland). We have not tested Cantonese: Eleven v4 lists it, and some community voices are tagged Cantonese, but our designed voices are Mandarin and the studio has no language switch. Try one short line before you commit a project to it.
English script, Mandarin audio
Text to speech does not translate: an English script read by a Mandarin voice comes out in English with a Chinese accent. Translate first and paste the Chinese. Two tools translate for you: video dubbing re-voices a video in Mandarin, and the audio drama studio can adapt a full-cast script into Chinese with the same cast.
Hear long Mandarin
All 534 episodes in the theater, 357 of them in Mandarin, were rendered on Eleven v4 with these voices. Zhu Ziqing's essay Father's Back, read in full:
Lu Xun's Kong Yiji with six voices:
What Chinese text to speech costs
Chinese is billed per character, and one Chinese character carries much more speech than one English letter: about 250 characters make a minute of Mandarin, against about 850 for English.
| Script | Characters | Credits | Audio |
|---|---|---|---|
| A short-video script | ~250 | 18 | ~1 minute |
| A 10-minute lesson | ~2,500 | 175 | ~10 minutes |
| One generation at the limit | 5,000 | 350 | ~20 minutes |
| A long-form job at the limit | 50,000 | 3,500 | ~3.3 hours |
Rate: 7 credits per 100 characters (4 on the faster Flash model). On monthly plans at list price, a minute of Mandarin comes to about $0.36–0.58 ($0.15–0.23 in a new account's first month). The trial pack is 150 credits, about 2,100 characters or 8 minutes, for $3.90 in the first 24 hours after sign-up (list $9.90), and its credits never expire. A first order is fully refunded within 7 days if you have used 150 credits or fewer. The audio you generate is yours to use commercially. See pricing.
FAQ
What is the most natural Chinese text to speech voice? It depends on the job. For news and explainers, News Anchor and Yun; for stories and history, Old Storyteller; for courses, Composed Lady; for ads and short videos, Gentle Healer; for romance and short drama, Domineering CEO and Ice Queen. All 15 are native Mandarin voices on Eleven v4; eight play on this page, and every one has its own voice page with samples.
Is this Chinese text to speech free? Listening is free: the samples, every voice preview and the 357 Mandarin episodes in the theater play without an account. Generating costs 7 credits per 100 characters, and 1,000 Chinese characters make about 4 minutes of audio. The one-time trial pack has 150 credits (about 2,100 characters, roughly 8 minutes of Mandarin, never expires) at a $3.90 new-user price in the first 24 hours after sign-up (list $9.90).
Does it support Cantonese? We have not tested it. Eleven v4 lists Cantonese among its languages and the community voice library has voices tagged Cantonese, but our 15 Mandarin voices were designed for Mandarin and the studio has no language switch. Try one short line before relying on it.
Simplified or Traditional characters? You can paste either. The voices speak Mainland Mandarin, so expect Mainland readings and accent even for Traditional text; Taiwanese Mandarin is not a native accent in our library.
How do I fix a character it reads wrong? In the text-to-speech tab, add the word once to a pronunciation dictionary with a homophone respelling (重庆 → 虫庆) and tick it; every later generation uses it, and the respelling never shows in the subtitles. Dictionaries work only in that tab (up to 5,000 characters per generation). In long-form and full-cast audio drama, type the homophone straight into the text instead; it will then show in the subtitles.
Can it turn English text into Chinese speech? Not by itself: text to speech reads the language you write, so translate the script first. Video dubbing and the audio drama studio can translate for you: dubbing re-voices a video in another language, and the drama studio adapts a full-cast script into Chinese with the same cast.