What the work involves

You receive German audio — recordings that may include regional accents, overlapping speech, background noise, code-switching into English, or domain jargon — and produce transcripts that hold up as training data. That means fidelity to what was actually said rather than a tidied-up version: filled pauses, false starts, dialect forms and self-corrections are usually preserved or tagged, not silently smoothed. Alongside verbatim work you'll apply the project's annotation layer: speaker IDs across turn changes, timestamps at the required granularity, and tags for inaudible segments, crosstalk or non-speech events.

A second stream of work is review. You'll be handed transcripts produced by other contributors (or by an ASR model) and asked to correct them against the audio and the style guide, flagging systematic errors rather than just patching individual lines. Projects often shift guidelines mid-flight, so the ability to re-read a spec, notice what changed, and apply it consistently from that point on matters more than raw typing speed.

What the platform screens for

  • Actual German depth, probed in follow-up: dialect and register recognition, Swiss/Austrian variants, orthography decisions (ß vs. ss, compound spelling, Anglicisms), and how you handle speech that is grammatically broken but perfectly intelligible.
  • Guideline discipline — whether you can state a rule you disagreed with and still applied, and how you escalate ambiguity instead of quietly inventing a convention.
  • Transcription practice: tooling you've used, timestamp conventions, how you mark uncertainty, turnaround you can sustain.
  • Confidentiality habits, since audio frequently contains personal information.

Logistics

Fully remote and asynchronous, contractor engagement. Work arrives in batches with deadlines rather than fixed shifts; contributors who can commit to a predictable weekly block tend to get steadier allocation. You'll need a quiet listening environment, good headphones, and a stable connection. No prior AI experience is expected — the screening is about your language work, not machine learning.