Chinese is written in characters with no spaces between words, and the transcript is written the same way, with ,and 。 for commas and full stops. It is not pinyin and it carries no tone marks. For anyone who reads Chinese this is simply normal text. For a learner it means the transcript can be pasted into a dictionary or a reading tool that adds pinyin, since the output is real Chinese script and not a romanised approximation.
Without word boundaries, the timestamp on each segment is the practical way to move between the text and the recording. Browser search also behaves differently: looking for a two-character word will find it inside longer expressions as well. Numbers, dates and amounts are usually written in Arabic digits, which is worth checking where a figure matters.
Mandarin is written in simplified characters in mainland China, Singapore and Malaysia, and in traditional characters in Taiwan, Hong Kong and Macau. There is no setting here to choose between them. The model writes what it judges fits the recording, which in most cases means simplified characters, and a recording from Taiwan may come back in either set or occasionally with a few characters of the other set mixed in.
If you need one set consistently, convert the finished text with any simplified-to-traditional converter; the conversion is mechanical and takes a second. It does not adjust regional vocabulary, so software stays 软件 turned into 軟件 and does not become the Taiwanese 軟體. Read through the result if the text is going to be published for a specific region.
Mandarin has few distinct syllables and a great many characters that share them. Shì alone can be 是, 事, 市, 式, 试 and dozens more. A speech model chooses the character from the surrounding sentence, and in clear lectures, news-style speech and ordinary conversation it mostly chooses well. Names are the case that listening cannot settle: a given name can be written with many different characters that sound identical, and nothing in the audio says which. Check personal names, company names and specialist terms against a participant list or the slides.
Standard Mandarin, pǔtōnghuà, is recognised best, and Taiwanese Mandarin transcribes well too. Speakers with a strong regional accent, where zh and z or n and l merge, need more correction. Cantonese, Shanghainese, Hokkien and other Chinese languages are a different matter: they are separate spoken languages, results for them are noticeably weaker, and spoken Cantonese may come back rewritten into standard written Chinese. We do not promise good results outside Mandarin.
Chinese files uploaded here are usually one to two hours long: university lectures and online courses, recorded meetings, interviews, and talks, podcasts and long videos from YouTube that language learners transcribe to read along. English terms inside a Chinese sentence, common in technical and business talk, are normally left in Latin letters. Guests get 10 transcriptions of up to 3 hours each. The free allowance is 5 hours in total across those 10 recordings; a free account keeps the full transcripts. AI chat with the transcript, automatic summaries and PDF or Word export are in the paid plans, and subtitles can be translated into other languages.
Recordings are processed on our own servers with our own speech models and are not sent to a third-party speech API. For internal meetings and client interviews, that removes one step from the approval most companies require before audio can be uploaded anywhere.