How it works
Paste Chinese text, simplified or traditional, choose a voice and press Generate. The text is split into sentences, each sentence is turned into Mandarin pronunciation with the right tones, and the first sentence starts playing as soon as it is ready while the rest is generated behind it. When it is done, download MP3 or WAV audio, or an SRT subtitle file timed to the speech.
Everything runs on your own device. The first time you use it, the Chinese voice engine (about 100 MB) is downloaded once and kept by your browser; after that it starts quickly. Your text is never sent anywhere, which is why there is no character limit and no account.
Getting the tones and readings right
Chinese pronunciation depends on context, so the text is first split into words, and each word gets its reading from a large dictionary. Characters with several readings are read by meaning: 银行 is yínháng, but 行长 in 银行的行长 is hángzhǎng, and 长大 is zhǎngdà while 很长 is hěn cháng. Tone changes are applied the way people speak: 不 before a fourth tone becomes bú, 一 changes with the next syllable, two third tones in a row become second plus third, and particles such as 的, 了 and 着 are neutral.
Numbers are read the way a person would say them: years digit by digit (2025年 as 二零二五年), 2500 as 两千五百, phone numbers digit by digit with 1 read as 幺, 37.5% as 百分之三十七点五 and 9:30 as 九点三十分. English words inside Chinese text, such as brand names or APP, are read with English pronunciation.
Voices
There are 100 Mandarin voices, 55 female and 45 male, recorded from a professional speech dataset. Twelve are shown first; the rest are under More voices. Every voice has a listen button that works before anything is downloaded, so you can pick by ear. The voices speak standard Mandarin as heard in mainland China.
Speed goes from 0.7× to 1.5×. Long sentences are automatically read a little more slowly so they never sound rushed. The voice and speed you pick are remembered on this device.
Traditional characters
Traditional Chinese text is read too: it is converted to simplified characters internally before pronunciation, including common Taiwan wording such as 軟體 and 網路, so the reading is correct. The audio is Mandarin with a mainland accent. A few words that are pronounced differently in Taiwan, such as 垃圾 or 和 used as a conjunction, follow the mainland reading.
Fix one sentence without redoing everything
After generating, the Parts list shows each sentence group with its length. Click one to play from there. If a character is read wrong or a sentence needs rewording, edit it in the list and press Save & regenerate. Only that part is generated again. For a wrong reading, the quickest fix is to replace the character with another that has the reading you want.
Speed on different devices
On computers with a modern graphics chip, Mandarin audio is generated several times faster than it plays, so playback starts within about a second. Browsers without that support generate more slowly than real time: the tool waits until enough audio is ready to play without stopping and tells you how long that is. For the best experience use a current version of Chrome or Edge on a computer.