$42/hr
The role in one line
A native French speaker audits AI training audio data to ensure transcriptions and word-level timings are accurate.
Written by Training Turk from the public listing; it may be incomplete or out of date. Read the full posting on Mercor.
What you would do
- Listen to recordings and compare them to transcriptions to verify accuracy, using systematic error codes for identified problems.
- Examine word-level timing boundaries set by automated systems and correct start-end timestamps when they misalign with actual pronunciation.
- Apply a fixed error classification scheme consistently, rating each task as pass or fail based on quality criteria.
- Write brief explanations for every judgment, flagging critical errors that explain a failure verdict.
Who they are looking for
- Native or near-native French from France, acquired through birth and childhood or achieved through five-plus years of continuous residence.
- Strong English reading and writing ability to comprehend rulebooks, error codes, and audit guidelines.
- Disciplined approach to systematic rubric application rather than relying on instinct or changing standards.
- Careful listening skill to distinguish similar sounds, identify word boundaries, and spot transcriber mistakes.
What the interview is likely to probe
1.Transcription accuracy judgment
Your core responsibility is determining whether transcriptions match reality; interviewers assess whether you hear errors others might miss.
Expect something like: “You hear a French phrase where the speaker pronounces a word ambiguously, and the transcriber chose one interpretation. How do you decide whether the transcription is accurate or erroneous, and what listen-again strategy do you use?”
2.Error code selection and completeness
Systematic feedback requires identifying every applicable error, not just the most obvious one; this trains models on nuance.
Expect something like: “In one segment you identify three different error types. How do you decide whether to tag all three separately or combine them, and how would your rationale differ from a colleague's?”
3.Timing boundary precision
Word-level alignments must be exact for voice AI training; interviewers verify you can detect misalignment at sub-second precision.
Expect something like: “A word starts at 2.34 seconds in the automated alignment but your ear tells you it begins at 2.29. How precise are you about this kind of boundary, and what acoustic cues help you judge?”
4.Distinguishing error from ambiguity
Not every discrepancy between transcription and audio is an error; genuine ambiguity should be handled differently.
Expect something like: “The speaker's pronunciation is unclear enough that both the transcriber's choice and an alternative interpretation seem defensible. How do you make a judgment in this case and what do you document?”
5.Consistency across many judgments
Models learn from consistent patterns; random judgment variation undermines training quality.
Expect something like: “You're 50 tasks into the work and encounter a category you initially rated strictly. How do you handle it now, and how do you maintain consistency without over-rigidity?”
Exercise you may get
Listen to five audio clips with transcriptions and word-level timings already assigned. For each, determine pass-fail status, apply appropriate error codes, verify or correct timing boundaries, and write clear judgments.
How to prepare
- Review the complete error code system and examples to internalize when each code applies.
- Practice listening to samples of the specific French dialect or region to develop your ear before production work.
- Study a calibration rubric or examples of high-quality audit judgments to understand what professional standard looks like.
- Reflect on your own listening biases and strategies for catching errors you might naturally miss.
Facts
- Pay
- $42/hr
- Commitment
- hourly
- Hours
- 10 per week
- Work arrangement
- remote · Remote
- Posted
- 9/18/2026