Cantonese Transcription Expert

$20–35/hr

rate as listed

Workload: ~15h/week
Engagement: Project-based
Location: Worldwide
Field: Languages

Transcribe and time-stamp Cantonese and English audio for an AI training dataset, and correct machine-generated transcripts against project style rules. Work is remote contractor work paid hourly, with output judged on accuracy of colloquial Cantonese, speaker labelling, and formatting consistency.

What the work actually involves

You receive audio — conversational, sometimes multi-speaker, sometimes code-switched between Cantonese and English — and produce a clean, time-stamped transcript that follows a written style guide. A large share of tasks are post-editing: an ASR system has already produced a draft, and your job is to catch what it got wrong. That typically means colloquial particles, sentence-final 嘅/㗎/囉, numbers and dates spoken in Cantonese, English loanwords rendered phonetically, overlapping speech, and proper nouns the model has never seen. You will also flag ambiguous segments rather than guess, and occasionally write short notes on why a passage is unclear.

The orthography question matters here. Cantonese transcription projects differ on whether to write spoken Cantonese in written Cantonese characters (係, 唔, 嘅, 咗) or normalise to Standard Written Chinese, and whether output is Traditional or Simplified. Strong contributors ask which convention applies before starting, then apply it without drifting across a long file.

What the screen looks for

micro1's screening is AI-led and conversational. Expect it to test whether your Cantonese is genuinely native — including regional variation between Hong Kong, Guangzhou and Macau usage — whether you have handled real transcription volume with real deadlines, and whether you can articulate a convention rather than just apply it by instinct. Being able to say "I use square brackets for inaudible with a timestamp, and I don't clean up disfluencies unless the guide says verbatim-lite" reads very differently from "I'm very detail-oriented." Prior AI work is not required and is not asked for.

Logistics

Remote, contractor, asynchronous; tasks are claimed and returned against milestones rather than worked on a fixed shift.
Observed pay band $20–35/hr, typically varying with audio difficulty and whether you are transcribing from scratch or reviewing. Rates are as observed on the platform, not guaranteed.
Volume fluctuates with client batches — treat it as supplemental rather than steady full-time hours.
You need your own quiet workspace, good headphones, reliable internet, and comfortable typing in Chinese input (Jyutping, Cangjie, or handwriting/pinyin as suits you).

What they look for

Native Cantonese proficiency, including Hong Kong and Guangdong regional variation
Professional audio transcription experience with Cantonese and English content
Accurate time-stamping and adherence to a written transcription style guide
Chinese typing in both Traditional and Simplified script
Post-editing of ASR output — catching machine errors rather than re-typing
English comprehension sufficient for code-switched and bilingual audio

Before you apply

Would you pass this screen?

Drawn from what this listing actually requires. Answered in your browser — nothing is sent to us, and we do not store your answers.

Is Cantonese your native language?
Native Cantonese · required
Have you been paid for audio transcription work involving Cantonese or English content?
Transcription experience · required
Can you type in both Traditional and Simplified Chinese using a keyboard input method?
Chinese typing · required
Can you accurately transcribe English speech within bilingual Cantonese-English recordings?
English comprehension · required
Do you have reliable internet, quality headphones, and a quiet space for detailed audio work?
Equipment and workspace · required
Have you previously annotated or reviewed linguistic data for a technology, media, or research project?
Linguistic data work · preferred

0 of 6 answered

Included with this listing

Interview prep — 10 questions

Matched to what this platform's screening actually measures. Answer each out loud in 60–90 seconds: claim → concrete example → implication.

What it tests: Whether you have produced transcripts under someone else's style guide and deadline, as opposed to casual bilingual note-taking.
A strong answer: Name the content type (court, media subtitling, medical, call centre, research interviews), the language mix, an honest hours-or-minutes-per-week figure, and who set the quality standard you worked to.
Where people lose it: Answering with language ability instead of transcription output — 'I'm a native Cantonese speaker and fluent in English' describes millions of people and says nothing about whether you can deliver a formatted transcript.

Summary, eligibility guidance and screening preparation are prepared by Sidequest from the platform's posting. Check the source listing before applying.