Turn Mandarin Chinese speech into Chinese characters on your own device: drop a recording or a video and get a plain transcript or timed .srt subtitles. Chinese is the largest language in Whisper’s training after English, and Mandarin is the variety it handles best. Two things to know before you start. There is no switch between Simplified and Traditional characters, and the model can produce either. And because so many syllables share a sound, names and rare words can come back in the wrong characters. Nothing is uploaded, and it works offline after the first run.
How it works
- Drop a Chinese recording or video. Chinese is already selected as the spoken language.
- Use Accurate (≈77 MB) for strong regional accents, specialist lectures or noisy audio; Fast (≈41 MB) for a quick draft. Choose Subtitles under Output first if you want .srt captions.
- Press Transcribe, check names and terms against the audio, fix characters and punctuation in place, and download .txt, .srt or .vtt.
Questions
- Does it output Simplified or Traditional Chinese?
- It can produce either, and occasionally mixes the two in one transcript. Whisper has a single Chinese setting and learned from text in both scripts, so there is no switch here to force one. If you need a single script throughout, run the finished .txt or .srt through a Simplified–Traditional converter.
- Can it transcribe Cantonese?
- Not reliably. The tiny and base models used here have no Cantonese language of their own, so Cantonese goes through the Chinese setting, which is strongest on Mandarin. Expect more errors, and text that may follow standard written Chinese rather than written Cantonese characters. Mandarin spoken with a Cantonese accent transcribes like other accented Mandarin, best with the Accurate model.
- Does my file get uploaded?
- No. The models and all processing run inside your browser on your own device. You can verify this yourself: open your browser’s DevTools, watch the Network tab, and process a file: there are no uploads. After the first visit most tools also work fully offline.
- How accurate is it?
- The models are OpenAI’s Whisper (tiny and base tiers). On clear speech they are very usable; heavy accents, crosstalk and noisy rooms will show errors. The Accurate tier is noticeably better on difficult audio.
- Which languages work?
- Fifteen, picked in the Language setting: English, Spanish, French, German, Italian, Portuguese, Dutch, Russian, Polish, Turkish, Japanese, Korean, Chinese, Arabic and Hindi. The decoder is conditioned on the one you pick, so it transcribes that language rather than approximating it in English — which also means picking the wrong one returns confident nonsense. The weights understand 99 languages; these are the ones we measured as good enough to offer.
- How long does it take?
- Transcription runs on your processor, so speed depends on the machine: expect around a quarter of the recording’s length, and less on the Fast model than on Accurate.