Korean Audio Transcription Reviewer

About OpenTrain

OpenTrain is the leading platform for people building careers in AI training and data labeling. We connect contributors with impactful projects, help you build a unified portfolio of AI-training work, and support durable freelance careers in a fast-growing industry.

About AI training and this work

AI training (data labeling/annotation) is the human work that teaches models how to understand speech, text, images, and more. Contributors directly shape how modern AI systems behave by checking outputs, correcting errors, and setting quality standards.

This role focuses on speech and transcription quality: you will assess AI-generated transcripts, speaker diarization, timestamps, and related metadata so speech models improve on realistic, multi-speaker Korean audio.

The role

You will review AI-generated Korean transcripts and diarization for transcribed, multi-channel audio recordings. Work centers on transcript fidelity, speaker alignment, word- and turn-level timing, JSON metadata verification, and deciding whether outputs meet strict accuracy and WER targets.

This is a part-time contractor role (20+ hours/week) and is fully remote. Work is evaluation-focused: you will spot recognition, segmentation, alignment, and attribution errors in spontaneous, overlapping speech.

What you'll do day to day

• Compare AI transcripts against audio for verbatim accuracy including disfluencies, interruptions, and false starts.
• Validate diarization outputs: speaker labels, speaker turns, and channel alignment in multi-speaker recordings.
• Check word-level and turn-level timestamps and ensure timing logic holds for overlaps and rapid turn-taking.
• Read and verify JSON-formatted outputs and metadata tagging for correctness and completeness.
• Decide whether outputs meet strict accuracy thresholds and report when WER/quality targets are not met.

Requirements

Preserve the following skills and qualifications exactly as stated; we do not invent or assume additional requirements.

• Fluent or native Korean proficiency (required).
• Experience reviewing ASR or AI-generated transcription outputs.
• Ability to evaluate diarization, timestamps, and speaker attribution reliably.
• Strong ear for audio fidelity, including background noise, channel bleed, and clipping.
• Careful attention to detail for verbatim transcription and timestamp checks.

Helpful background

The following backgrounds make you especially competitive but are not strict requirements: linguistics, phonetics, transcription, or speech data QA experience, a deep understanding of natural, unnormalized speech patterns, and prior work with multi-speaker conversational audio.

• Familiarity with conversational dynamics in overlapping dialogue and turn-taking.
• Experience validating turn-level and word-level timestamps in complex audio.

How this work is structured

This is contract, part-time work through OpenTrain. You will work remotely and independently, choosing shifts that fit the 20+ hours/week expectation and following detailed annotation and QA guidelines provided with each assignment.

Tasks require careful review of audio and machine outputs; you will submit evaluations and pass/fail decisions against explicit accuracy and WER targets. No pay or rate is listed here; compensation and task assignment details are provided during onboarding.

• Employment type: Contractor, Part-time.
• Time expectation: 20+ hours/week.
• Language requirement: Korean (ko).
• Data type: Audio; label types: Transcription, Evaluation/Rating.