What the work actually involves

You receive French audio — interviews, calls, spontaneous speech, sometimes noisy or accented — alongside an ASR system's first pass at transcribing it. Your job is to bring that transcript to a defensible standard: fixing misrecognised words, resolving overlapping speech, marking inaudible segments honestly rather than guessing, and handling the things automated tools reliably get wrong in French — liaison and elision, homophone clusters (a/à, ce/se, ses/ces/c'est), verbal endings that sound identical (-é/-er/-ai), proper nouns, and code-switching into English or Arabic. On top of the text, you apply metadata: speaker turns, dialect or regional variety (Hexagonal, Québécois, Belgian, West African, Maghrebi), register, disfluencies, background events, and whatever tag set the customer's guideline specifies.

The annotation layer is where most of the judgment lives. Guidelines are rarely complete, and you will hit cases they did not anticipate — a speaker who slides between vous and tu mid-sentence, a regionalism that is not an error, a segment where the audio genuinely supports two readings. The expectation is that you apply the guideline consistently, flag the ambiguity rather than silently inventing a convention, and raise it with the customer's team so the guideline gets fixed for everyone.

What the screen looks for

micro1's screening is AI-led and conversational. It probes native-level command of spoken French under follow-up — expect to be asked about specific phonetic or orthographic traps, not asked to assert that your French is good. It also tests whether you have handled real transcription conventions before (verbatim versus clean-read, timestamping, speaker labelling) and whether you can describe a concrete case where you disagreed with a style guide or an automated output. Professional-level English matters because guidelines, tooling and team communication are in English. No AI or machine-learning background is required; the domain knowledge is the credential.

Logistics

  • Fully remote, contractor engagement, invoiced hourly
  • Largely asynchronous, with periodic syncs or written check-ins with the customer's team
  • Task volume fluctuates by project; treat it as flexible part-time rather than guaranteed hours
  • Requires a quiet working environment, reliable connectivity, and headphones good enough to resolve degraded audio
  • Pay band of $20–36/hr is as observed on this listing, not a guarantee; placement within it typically reflects demonstrated transcription experience and dialect coverage