Turn Chinese meetings, interviews and lectures into searchable text with speaker labels — Mandarin and Cantonese, Simplified or Traditional characters.
iOS and Android. 30 minutes free, no card.
Does SozAI transcribe Chinese?
SozAI transcribes Chinese audio and video into text for Mandarin and Cantonese, with automatic speaker labels and word-level timestamps. Output can be Simplified or Traditional characters, and exports to TXT, DOCX, PDF, SRT and VTT. Recordings are processed in EU data centers and encrypted at rest, and every account gets 30 minutes a month without a card.
中文, written the way it is actually written.
Specification
| Native name | 中文 |
|---|---|
| ISO 639-1 code | zh |
| Writing system | Han (Simplified and Traditional) |
| Text direction | Left to right |
| Speakers worldwide | ~1.1 billion |
| Varieties covered | Mandarin, Cantonese |
| Speaker labels | Yes — automatic diarization |
| Timestamps | Word level |
| Summaries | Yes, in the same language |
| Export formats | TXT, DOCX, PDF, SRT, VTT |
| Maximum file size | 500 MB |
| Free tier | 30 minutes per month, no card |
| Where data is stored | EU data centers, AES-256 at rest |
| Public transcripts in this language | 274 |
Workflow
Record straight into the app, upload a file up to 500 MB, or paste a YouTube link. SozAI reads m4a, mp3, wav, aac, mp4 and mov.
Choose Mandarin or Cantonese and your character set, or leave detection on. Recordings that mix Chinese with English technical terms stay in a single transcript.
The transcript arrives with speaker labels and word timestamps. Search it, ask it questions, generate a summary in Chinese, or export SRT and VTT subtitles.
Field notes
Mandarin distinguishes syllables by pitch contour, so mā, má, mǎ and mà are four unrelated words. A tone error is a word error, not a pronunciation quirk, and compressed phone audio is exactly where tone contours degrade. This is the single biggest driver of accuracy differences between clean and noisy Chinese recordings.
Chinese is written without word boundaries. Before the transcript can be searched or time-aligned, the system has to decide where words start and end, and different segmentations of the same characters can be equally grammatical with different meanings. Search behaviour in Chinese transcripts depends entirely on getting this right.
They are not mutually intelligible in speech, and Cantonese has its own vocabulary, grammar and written characters. Treating Cantonese as accented Mandarin produces confident nonsense, so it is a separate target. Recordings that switch between them within one meeting remain the hardest case.
The same spoken sentence is written differently in mainland China than in Taiwan or Hong Kong, and the mapping is not always one to one. Choosing the wrong character set produces a transcript that is technically readable and immediately marks the document as foreign to its audience.
From the archive
274 public Chinese transcripts so far — every one produced by the same engine described above. Read them in full, no account needed.
西安事变内幕 挽救共产党 | 还原一个真实的张学良 | 不忠不孝不义不仁 (中文字幕)#矮大紧#晓松奇谈#晓说… — Transcript![kubernetes начало [2025] — Transcript — Chinese transcript](https://i.ytimg.com/vi/NpFjx-v7S58/maxresdefault.jpg)
kubernetes начало [2025] — Transcript
🍅✨【全集】😱傅家只是隨手收養一個小女孩，誰知她身帶治癒花香🌸病弱少爺因她好轉，豪門大佬也接連登門！👑 #豪門… — Transcript
AI下一场战争，不是只拼模型 | AI竞争 | 算力独立 | 开放模型 | 机器人 | AI生物学 |…
Narasimha Story | Swami Aniruddha | — Transcript
【Podcast】EP72｜床上有哪些地雷和隱藏技巧？A片裡看起來很爽其實超不爽的NG行為？真的有高朝地圖這種… — TranscriptSozAI transcribes over 100 languages. These have a page of their own.
Answers
Short answers you can quote. Longer ones live in the sections above.
Yes, as a separate target rather than as an accent of Mandarin. The two are not mutually intelligible in speech and Cantonese has its own vocabulary and written forms, so selecting the right one matters more than it does for most dialect choices in other languages.
Yes. Both character sets are supported, and the choice affects the whole transcript. Because the mapping between them is not strictly one to one, picking the right one up front produces better output than converting afterwards.
Yes. Diarization runs on the audio rather than the language, so each speaker gets a label and every turn a timestamp. For 会议记录 this is what turns a recording into a document you can circulate rather than a wall of characters.
SozAI exports SRT and VTT in UTF-8 with word-level timing, so Chinese characters load correctly in YouTube, Premiere Pro, Final Cut Pro and DaVinci Resolve. Line breaking still deserves a human pass, since Chinese subtitle conventions differ from Latin-script ones.
Every account gets 30 minutes per month free without a card. Beyond that SozAI is a flat monthly subscription with a shared pool of minutes, and the price does not depend on the language. Current plans are listed on the pricing page.
Where to use it
Everything below works in Chinese, with the same speaker labels and word timestamps.
Transcribe your first Chinese recording
Record, upload a file or paste a link. Speakers, timestamps and a summary come back together.
Free on iOS and Android. No account required for YouTube transcripts.
Point your camera
Scan the code to open the download page on iOS or Android. 30 minutes of transcription free, no card.
