Description
Review pre-segmented audio recordings, verify word-level segment accuracy, and correct timestamp or transcription errors to support training data for speech-to-speech AI models.
Key Responsibilities
- Listen to short audio clips and review automatically generated word-level segments (start timestamp, end timestamp, transcription)
- Verify segment boundaries align with actual speech onset and offset
- Review start/end timestamps when boundaries are inaccurate
- Identify and flag tasks for discard per defined criteria (unintelligible audio, non-target language, sensitive content, no voice activity, etc.)
- Maintain consistency and accuracy across a detailed, evolving style guide
Requirements
- English fluency, including strong command of spoken/colloquial forms (contractions, filler words, informal speech)
- Prior experience with audio transcription, data annotation, or speech/NLP data labeling preferred
- Comfortable using annotation/labeling tools
- Ability to reference and apply a detailed style guide consistently
- Ability to work independently and adapt to periodic guideline updates