What the work involves

You work through queues of short Bengali audio clips. Some are human speech, some are synthesised or model-generated, and often you will not be told which. For each clip you judge whether the speaker sounds native, where the pronunciation drifts, whether intonation and prosody match natural Bengali patterns, and whether anything about the delivery marks it as non-native or machine-produced. Each rating is accompanied by a written justification in English, pointing at specific phonetic or prosodic evidence rather than a general impression.

The written half matters as much as the listening. Annotators who write "sounds unnatural" produce data nobody can act on, whereas "retroflex /ɖ/ realised as dental, vowel length collapsed in the second syllable, falling contour where a rise is expected in a yes/no question" is usable. Guidelines evolve, and you will be expected to raise ambiguous cases with project coordinators instead of silently guessing, particularly around dialect variation (Bangladeshi versus West Bengal varieties, Sylheti and Chittagonian influence, formal sadhu bhasha versus colloquial cholito bhasha).

What the platform screens for

micro1 runs an AI-led interview. Expect it to probe your Bengali background directly: where you learned it, which variety you speak, whether you can describe specific phonological features under follow-up questions. It will also test whether your English is strong enough to write evaluation notes a non-Bengali-speaking researcher can read. Claims of "native proficiency" get followed up, so be ready with specifics: tone sandhi, aspiration contrasts, schwa deletion, whatever you actually know. No AI experience is required; the screen is checking domain depth and whether you can apply a rubric consistently.

Logistics

  • Fully remote, contractor engagement, asynchronous task queues
  • Quiet listening environment and decent headphones are effectively required
  • Work volume comes in project batches; hours are flexible but deadlines on assigned batches are not
  • Pay band: $30–65/hr (observed, not guaranteed)