What the work involves
You receive batches of short Dutch audio clips produced by speech models and rate them against native-speaker standards. That means listening closely for vowel quality and diphthongs, consonant assimilation and final devoicing, word and sentence stress, intonation contours, and whether the output lands in a recognisable variety (Standard Dutch, a Flemish register, a regional accent) or drifts into something no native speaker would produce. Ratings alone are rarely enough: most tasks ask for a written justification in English naming the exact segment, word or syllable where the output broke down and why it sounds off.
Expect a rubric with defined scales, calibration examples, and occasional reversals when quality leads disagree with your call. Consistency across a batch matters as much as any single judgement, so you will be asked to apply the same threshold to clip 200 as you did to clip 2. Project update sessions and clarification threads are part of the job, and raising an ambiguous case early is treated as a contribution, not a complaint.
What the platform screens for
- Native-level Dutch, with the ability to describe differences between Netherlands and Belgian Dutch concretely rather than in general terms.
- English at B2 or above, since the feedback itself is usually written in English.
- The habit of phonetic and grammatical specificity: "the /ɣ/ is realised too softly" beats "sounds unnatural".
- Rubric discipline, including the willingness to score consistently when your personal taste differs from the guideline.
Logistics
Remote contractor work, largely asynchronous, with hours set around batch availability and your own schedule. Volume fluctuates with client demand, so this suits people who want flexible supplementary work rather than a guaranteed weekly load. A quiet environment and decent headphones are effectively required for audio work.
Pay band: $30–65/hr (observed, not guaranteed).