What the work actually involves

You receive batches of Korean audio — conversational, read, or spontaneous speech of varying recording quality — and produce verbatim transcripts that follow a client style guide. That guide dictates decisions you would otherwise make by instinct: whether to spell out numbers or use digits, how to mark disfluencies and filled pauses, whether to render English loanwords in Hangul or Latin script, how to tag overlapping speakers, unintelligible spans, and background noise. A second stream of work is quality assurance: reviewing another transcriber's output, correcting errors, and flagging where the guideline itself is ambiguous. You will also add metadata and contextual notes — speaker turns, dialect markers, emotional or acoustic context — that the model training pipeline consumes downstream.

Code-switching is the part most candidates underestimate. Korean media and casual speech mix English freely, and Konglish forms, English proper nouns spoken with Korean phonology, and partial English clauses all need consistent handling. Regional variation matters too: Gyeongsang and Jeolla speech, Jeju forms, and North Korean-influenced speech patterns appear in real corpora and are not always signposted in the file.

What the screen looks for

micro1's process is AI-led and conversational. It probes for native-level Korean comprehension across registers, actual transcription experience where accuracy was audited, and — most revealingly — your judgment when a style guide does not cover the case in front of you. Expect follow-ups that push on specifics: which tools you used, how your accuracy was measured, what you did with a five-second unintelligible stretch. General claims of fluency and attention to detail do not survive the second question. No AI background is required; the linguistic depth is the qualification.

Logistics

  • Fully remote, contractor engagement, asynchronous within assignment deadlines
  • Hours are flexible but volume-driven; many contributors work part-time alongside other commitments
  • Requires a quiet workspace, reliable internet, and good headphones — laptop speakers are not sufficient for disputed audio
  • Pay observed in the $20–36/hr range, varying with task type and QA responsibility; not guaranteed