Which language to pick for a recording from China
China has one official spoken standard, Putonghua, and many regional languages that are not mutually intelligible with it. The upload box needs to know which one is on the recording, because each option is written differently.
- Putonghua and regional Mandarin (Beijing, Northeast, Sichuan, Shandong and most of the north and southwest): keep Chinese selected.
- Cantonese from Guangzhou, Foshan or elsewhere in Guangdong and Guangxi: choose Cantonese. It is a separate option with its own page.
- Shanghainese and other Wu varieties, Hokkien and Teochew (Min), Hakka, Xiang and Gan: there is no separate option. Run with Chinese selected only if most of the talk is in Putonghua, and expect the local-language passages to be missed or rewritten.
- Tibetan and Mongolian appear in the language menu, but they have far less training data. Treat that output as a rough draft for a fluent speaker to check.
Regional Mandarin and code-switching between Putonghua and dialect
Many recordings made in China switch registers. A teacher in Chengdu lectures in Putonghua and tells a joke in Sichuanese. A family in Wuhan talks in local speech and switches to Putonghua for a grandchild. A factory meeting in Dongguan runs in Putonghua while two managers whisper in Cantonese.
With Chinese selected, the model writes everything in standard Mandarin characters. Southwestern and Northeastern Mandarin are close enough that the meaning usually survives, though local words such as 巴适 or 嘎哈 may come out as near-sounding standard words. Passages in a non-Mandarin language are where gaps and invented text appear. Mark those timestamps and fill them in by ear.
Auto-detect reads only the opening of a file, so a recording that starts with a Cantonese greeting can be labelled Cantonese for the whole hour. Choosing the language yourself avoids that.
Sound conditions in China add their own problems. Lecture halls often feed the teacher through a loud PA system with echo, and restaurant or street recordings carry heavy background noise. Record as close to the speaker as you can, and upload the original file rather than a compressed copy sent through a chat app.
Simplified characters and mainland formatting
Mainland documents use Simplified characters, which is what Mandarin transcripts normally come out in, so a transcript for a mainland reader rarely needs conversion. If the same text will also go to readers in Hong Kong or Taiwan, convert the exported file with a separate tool afterwards and recheck the names.
Before a transcript goes into a report or a paper, check three things: the punctuation style your publisher expects, whether large numbers should be written with 万 and 亿 or as digits, and the characters of every personal and place name. A county name heard once in a fast interview is the most likely word to be wrong.
Typical recordings from China
These are the files people most often bring from China. Upload audio or video directly. Audio is pulled out of MP4, MOV and other video formats on our side.
- University lectures and seminar recordings, turned into study notes.
- Fieldwork interviews by sociologists, anthropologists and journalists, often recorded in villages where the speaker mixes Putonghua with local speech.
- Business meetings and supplier calls, kept as a searchable record.
- Family recordings of grandparents, where the language choice matters most.
- Videos from Chinese platforms. Links from some sites need a login and will not import, so download the file and upload it instead.
After the transcript is ready
Click a line to hear it and correct characters in place. Export to TXT, DOCX, PDF, SRT, VTT or CSV, with timestamps and speaker names on or off. Speaker labels are available on paid plans or with a Minute Pack.
The text stays in Chinese. For an English reading, use the Google Translate link on the transcript, or the AI prompt export to paste the text into ChatGPT, Claude or Gemini with a ready-made prompt.