Chinese audio and video to text

Mandarin, Cantonese and other varieties

Standard Mandarin is the safest input, including the accents of Beijing, Taiwan, Singapore and Malaysia, though heavy regional accents lower quality. Cantonese, Hokkien, Shanghainese and other varieties are much weaker, and speech in them may come back in a written form you did not expect. A speaker who mixes Mandarin with English product names or a local variety will need the most proofreading.

Simplified or Traditional characters

Chinese is written without spaces, with full-width punctuation such as ，。？. The model decides between 简体 and 繁體 output, and you cannot set it. Both can appear in one transcript. If you need one script only, export the text and run it through a separate conversion tool, then read it again. Homophones and personal names are the usual errors.

What speakers of Chinese use it for

University lectures and online courses, meeting notes for teams that work in Chinese, interviews for research or journalism, and captions for videos posted to Bilibili, YouTube or short-video apps. People also transcribe voice messages they have saved as audio files, or family recordings, to search them or share them as text.

Which model works best for Chinese

Chinese is one of the languages Whisper-family models have seen the most speech in, so the small model can give a reasonable first draft from clear audio. That is not a promise of accuracy: it still slips on names, numbers, jargon and overlapping or noisy speech, so proof-read anything you will publish or rely on.

On tryscriptorium.com the no-account preview and regular accounts run the small model (the larger medium and large-v3 models run when you self-host Scriptorium).

Chinese has far more training data behind it than most languages, so clear Mandarin from one speaker is a reasonable case even on the small model. Names, place names and technical terms still need a proofread. For Cantonese or dense mixed-language speech, self-hosting with large-v3 gives the best chance.

Subtitles and captions in Chinese

Keep Chinese subtitle lines short, roughly 15 to 20 characters, because there are no spaces to break on and long lines are hard to read at speed. Split at natural phrase boundaries, not mid-word. SRT and VTT are UTF-8, so Chinese characters display correctly in VLC, YouTube and most editors. Spoken Chinese is dense, so shorten lines rather than speed up the reading.

The uploader is set to Chinese

The upload box on this page transcribes as Chinese, so the recogniser does not have to guess the language from the first seconds. If the recording is in another language or switches languages, use Audio to text, which detects the language automatically. Automatic transcription is a draft and no accuracy figure is promised: check names, numbers and anything you will rely on.