- Develop, refine, and apply audio quality guidelines for speech and voice datasets.
- Review audio files against technical, linguistic, and task-specific standards, making clear approval, rejection, or revision decisions.
- Identify audio and annotation issues such as background noise, clipping, distortion, plosives, echo, low signal, segmentation errors, transcript mismatches, and speaker-label inconsistencies.
- Perform annotation and QA tasks, including transcription, timestamp validation, VAD/segmentation, diarization, pronunciation checks, and metadata review.
- Record speech based on provided scripts and performance guidelines, delivering natural, high-quality, specification-compliant audio.
- Document edge cases, update review rubrics, and improve internal SOPs and quality standards.
- Collaborate with research, ML, and operations teams to translate model requirements into data specifications and evaluation criteria.
- Ensure consistency and integrity across audio files, transcripts, annotations, and associated metadata.
Who we're looking for
The ideal candidate has direct experience working with audio AI datasets and understands what makes speech data effective for model training. You have a strong ear for audio quality, are comfortable applying annotation standards, and can consistently produce and evaluate high-quality recordings.
- Direct experience working with audio AI training datasets or evaluation workflows.
- Hands-on experience with TTS, ASR, transcription, speech-to-speech, or related voice AI systems.
- Experience developing or applying audio quality standards in production environments.
- Experience with speech annotation tasks such as transcription, timestamp QA, VAD/segmentation, and diarization.
- Strong auditory judgment with the ability to consistently identify subtle audio quality issues.
- Ability to produce high-quality recordings in a controlled, quiet environment using professional or near-professional equipment.
- Strong written communication skills with the ability to provide clear, actionable feedback.
- High attention to detail and sound judgment when evaluating edge cases.
- Comfort working with structured data formats such as spreadsheets, CSV, or JSON.