Transcribe Chinese audio to text, free

What makes Chinese hard to transcribe

Written Chinese has no spaces between words, so accuracy is measured in character error rate rather than word error rate. Mandarin also carries four tones plus a neutral tone, and homophone density is extreme — dozens of distinct characters can share one pronunciation, meaning the model has to pick from meaning, not sound.

What your Chinese transcript looks like

Output is a continuous character stream with punctuation, matching normal Chinese typography. Simplified characters are produced by default.

[00:00 → 00:04] 大家好，今天的会议现在开始。

[00:04 → 00:08] 首先我们来看上个月的数据。

An illustration of how Chinese transcripts are formatted — script, punctuation and timestamps. Not the output of any particular recording.

How accurate to expect

Mandarin from broadcast, lecture and podcast sources transcribes well. Accuracy falls on regional accents from outside the Beijing standard, and on recordings where speakers switch into a local dialect.

What usually needs fixing

1. 1 Homophone selection — the sounds are right but the character chosen carries a different meaning, most visibly in names.
2. 2 Company and brand names, where the correct characters are a business decision rather than a phonetic one.
3. 3 Numbers and measure words running together, since Chinese counts with a classifier attached to the number.

Getting better results

• → Say numbers and dates clearly — Chinese number words are short and easily merged with neighbouring characters.
• → Recordings that mix Mandarin with English product names transcribe fine; you do not need to separate them.

Who transcribes Chinese