Mandarin, Putonghua, Guoyu and Huayu
Mandarin is the English name for the family of northern Chinese varieties, and for the standard language built on the Beijing pronunciation. The standard goes by different names depending on where you are: Putonghua (普通话) in mainland China, Guoyu (国语) in Taiwan, and Huayu (华语) in Singapore and Malaysia. All of them are Mandarin.
In the upload box, Mandarin is covered by the Chinese option, code zh. This page selects it for you. Cantonese is a different spoken language with its own option, so a Hong Kong or Guangzhou recording in Cantonese should not be uploaded as Chinese.
Mandarin is one of the languages the speech model knows best. Clear speech gives a strong draft, but you still need to check names, numbers and terms before you publish or quote.
Mainland, Taiwan and Singapore Mandarin
The three standards share grammar and most vocabulary, yet everyday words differ. A video is 视频 on the mainland and often 影片 in Taiwan. Software is 软件 or 軟體. A taxi is 出租车, 计程车 or, in Singapore, 德士. The model writes the word the speaker used, but it may occasionally swap in the more common mainland term, so watch for this in Taiwanese and Singaporean recordings.
Taiwanese Mandarin tends to have softer retroflex sounds and fewer erhua endings, and Singapore Mandarin often carries English words and particles such as lah or leh. Code-switching with English is common in all three places. Short English words usually stay in Latin letters inside the Chinese text.
Transcripts normally come out in Simplified characters, whichever variety is spoken. If readers in Taiwan expect Traditional characters, convert the exported text with a tool you trust and recheck personal names afterwards.
Where Mandarin transcripts need a second look
Some errors come from the language itself rather than the audio. These are the ones that catch people out most often.
- 他, 她 and 它 are all pronounced tā. The model has to guess from context, so a woman may be written as 他 or the other way around. Check every tā in interviews about a specific person.
- 的, 得 and 地 all sound like de in fast speech. Formal documents expect the right one in each grammatical slot, and the draft does not always pick it.
- Particles and fillers such as 啊, 吧, 呢, 那个 and 就是 may be kept, dropped or merged. Decide whether you want a verbatim or a clean transcript and edit toward that.
- Numbers can be digits or Chinese numerals, and large amounts said with 万 or 亿 are easy to misread. Compare figures with the audio.
Uploading Mandarin recordings
Typical files are WeChat or WhatsApp voice messages saved as audio, Zoom or Tencent Meeting recordings, lectures, Bilibili or YouTube videos, and interviews for research. Upload them as they are; audio is pulled out of video automatically. For a YouTube or TikTok link, paste the link instead, though some sites need a login and won't work, in which case upload the file.
Keep Chinese selected even if a meeting starts with an English greeting. Auto-detect listens to the opening seconds and can label the whole file as English, which gives you a page of nonsense.
After the transcript is ready
Play the recording and follow the highlighted line, then click any line to hear it again and fix the characters in place. Export TXT or DOCX for notes and quotes, SRT or VTT for Chinese subtitles, or CSV for coding interviews. Speaker labels are on paid plans or with a Minute Pack.
There is no pinyin and no built-in translation. To read the content in English, use the Google Translate link on the transcript, or copy it with the AI prompt export into ChatGPT, Claude or Gemini and ask for a summary.