What the work involves

You receive audio files — call recordings, conversational clips, read speech, sometimes media excerpts — and produce verbatim or lightly cleaned Cantonese transcripts against a written style guide. In practice that guide is where most of the difficulty lives: whether to write colloquial Cantonese characters (嘅, 咗, 唔) or standard written Chinese, how to handle English words dropped mid-sentence, how to tag laughter, overlap, false starts, and inaudible spans. Tasks often include a second layer: reviewing another transcriber's output, correcting it, and leaving a short note explaining what was wrong and why.

A meaningful share of the work is judgment rather than typing. You will hit clips where the speaker's Cantonese is accented by Mandarin or a regional variety, where two speakers talk over each other, or where a proper noun is audible but ambiguous. The platform cares less about your typing speed than about whether you mark uncertainty honestly instead of guessing plausibly — a confidently wrong transcript is more damaging to a training set than a flagged one.

What the platform screens for

  • Native or near-native Cantonese listening comprehension, including colloquial and fast conversational registers
  • Ability to type Cantonese-specific characters fluently (Jyutping, Cangjie, or handwriting input all fine)
  • Consistency: applying the same convention across a long file, not drifting halfway through
  • English comprehension strong enough to follow written guidelines and code-switched speech
  • Prior transcription, subtitling, captioning, court/medical transcription, or linguistics work is a strong signal

micro1's screen is AI-led: an interview covering your background and Cantonese depth, usually plus a short sample transcription task. Expect follow-up questions that dig into specific choices you made.

Logistics

Fully remote and asynchronous — you claim tasks from a queue and work to a deadline rather than fixed hours. Volume fluctuates by project; contributors commonly report 10–25 hours a week available when a project is active, with quieter gaps between. You need a quiet environment, reliable internet, and good headphones; some projects require a specific transcription tool in-browser. Pay is hourly or per-audio-minute depending on the project, in the $20–35/hr observed range, with review and QA tasks generally at the upper end. Nothing here is guaranteed volume.