What the work actually involves
You receive batches of short audio clips of synthesised or model-generated Indonesian and rate them against a rubric. The core question is nativeness: does this sound like a person from a real speech community, or does it sound like a machine imitating one? That means catching flattened prosody, English-influenced vowel length, misplaced stress, phonemes that drift toward Malay, particles like sih, dong, kok used in the wrong register, and formal bahasa baku delivered in a context that would call for casual Jakarta speech. You then write it up — clip reference, what you heard, why it's wrong, what a native speaker would have produced instead. Some tasks include a verbal walkthrough or a live session with a project coordinator.
What the platform screens for
micro1's screen is AI-led and conversational, and it is looking for evidence rather than claims. Expect to be asked to describe specific phonological or pragmatic features of Indonesian and then pushed one level deeper on your answer — naming a dialect is not enough, you'll be asked what marks it audibly. Your English is assessed live too (the listing sets B2 as the floor), because the deliverable is written feedback in English about Indonesian. The other thing being measured is calibration: whether you can distinguish "unnatural" from "unfamiliar to me," and whether you follow a rubric you disagree with instead of substituting your own taste.
Logistics
- Remote contractor engagement, no location requirement stated beyond remote work
- Largely async batch work with scheduled coordinator check-ins; timezone overlap is occasionally needed for those
- Volume moves with client demand — expect uneven weeks rather than a fixed schedule
- Requires quiet listening conditions and decent headphones; laptop speakers will not surface the artefacts you're being paid to catch
- No prior AI or annotation experience required