Sonic Audit Specialist – Japanese at Mercor

Sonic Audit Specialist (Japanese) | $37.50/hr | Worldwide Remote

Mercor is seeking native and near-native Japanese speakers to audit AI training data for Amazon's Sonic audio collections. This is careful, detail-oriented listening work — not high-volume production — requiring strong rubric discipline and clear written reasoning in English.

About the Role

You will work across two Amazon Sonic audit collections:

• Multilingual Transcription Audit: Listen to recordings of human or agent speech, evaluate annotator-produced transcriptions for accuracy, apply a defined set of error codes, assign a pass/fail verdict, and write a concise rationale for each decision.
• Forced Alignment Audit: Review word-level audio segments produced by an automated system. Verify that start and end timestamps correctly match spoken words, confirm the reference transcription matches the spoken form, and correct any machine errors by adjusting boundaries, splitting or merging segments, or adding/removing segments as needed.

Each task runs at 5.00 audit hours. Every judgment requires a written explanation, making this a precision-focused role.

Who We're Looking For

• Native or near-native Japanese speaker (born/raised in Japan or fully fluent with 5+ years of residence) — this is a hard requirement.
• Strong English reading and writing skills — all conventions, error codes, and rationale are written in English.
• Disciplined rubric application — ability to apply written standards consistently across hundreds of judgments.
• Careful listening skills — able to distinguish similar sounds, identify word boundaries in continuous speech, and differentiate acoustic ambiguity from clear annotator error.

Preferred (Not Required)

• Prior experience in transcription, subtitling, localization, linguistic annotation, or QA
• Background in AI data annotation or model evaluation
• Phonetics or linguistics training
• Experience with audio editing or waveform tools

What to Expect

You will listen, judge, and submit. When errors are found, you will tag all applicable error codes and explain clearly why a task passed or failed — leading with critical errors. A short calibration phase is required before production work begins.