Chinese audio to text for interviews, lectures and family recordings
People outside China often have hours of Mandarin audio and no quick way to search it. A researcher records interviews in Chinese for a thesis. A journalist has a phone call with a source in Shanghai. A family has recordings of grandparents telling stories. A company has client calls with partners in Taiwan or Singapore. Typing these out by hand takes several times the length of the recording.
RocketScribe gives you a first draft in a few minutes. The transcript is split into paragraphs with timestamps. A synced player highlights the line being spoken, so you can listen and correct at the same time. Click any line to jump to that point in the audio.
What the Chinese transcript looks like
Mandarin speech comes out in Chinese characters, normally Simplified. There is no Traditional-character conversion in RocketScribe, so if you need Traditional characters for readers in Taiwan or Hong Kong, run the exported text through a conversion tool you trust, then check names, which converters sometimes get wrong.
A few things are worth knowing before you edit:
- Punctuation is added by the model and is sometimes lighter than you would write by hand. Long spoken sentences may need an extra comma or full stop.
- Numbers can appear as Arabic digits or as Chinese numerals, depending on how they were said and the context. Standardise them if your document needs one style.
- Homophones are the most common error. Mandarin has many words that sound alike, so personal names, company names and technical terms can come out with the right sound but the wrong characters. Search the transcript for each important name and fix it once you know the right form.
- English words spoken inside Chinese sentences usually stay in Latin letters. If your recordings switch languages a lot, see the Chinese–English mixed transcription page.
Accents and recording conditions
Mandarin is one of the languages the speech model has seen the most training data for, and clear Standard Mandarin transcribes well. Speakers from different regions add variation: northern speakers may use a lot of erhua, southern speakers may merge sounds such as zh and z, and Taiwanese Mandarin has its own vocabulary and rhythm. These usually still work, but expect more corrections as the accent gets stronger.
Regional languages are a different matter. Cantonese has its own option and its own page. Shanghainese, Hokkien and other Chinese languages are not offered as separate options, and speech in them will not transcribe reliably as Mandarin.
Audio quality matters as much as accent. A phone placed close to the speaker beats a laptop across the room. Background music, a busy restaurant or two people talking over each other will reduce accuracy in any language, and Chinese is no exception.
Set the language or let it detect
This page preselects Chinese, which is the safest choice for a Mandarin file. Auto-detect looks at the start of the recording, so it can guess wrong on very short clips, on recordings that begin with a few sentences in English, or on Cantonese speech. If the first lines of a transcript come out in the wrong language, run it again with Chinese chosen. You can change the language in the upload box before each file.
Using the transcript afterwards
Export to TXT or DOCX for reading and quoting, SRT or VTT for subtitles on a video, or CSV if you want to code interview segments in a spreadsheet. All of these keep every Chinese character. The direct PDF option supports Western European characters only, so for a Chinese transcript it opens a print-ready page that you save as PDF from your browser.
If you do not read Chinese, the Google Translate link opens the transcript in Google Translate so you can follow it in English. The AI prompt export copies the transcript together with a prompt you can paste into ChatGPT or Claude to summarise it or translate it. On paid plans, speaker identification labels who said what, which helps a lot with interviews and meetings.