Audio Data Labelling Specialist

**Audio Stem Labeller – Instrument Classification** **About the role** We’re building an audio classification system that identifies instruments in real-world music stems. We’re looking for someone with a strong ear for music production to help us label audio stems accurately by instrument. This is not a coding or machine learning role. The work is focused on listening carefully, identifying the instruments present, and applying consistent labels to messy real-world audio files. The stems may include clean recordings, processed sounds, layered parts, effects, pitch-shifted material, distorted sounds, MIDI instruments, sample-library sounds, and edge cases where the instrument is not immediately obvious. **What you’ll be doing** * Listening to individual audio stems and identifying the main instrument or sound source * Labelling stems using our agreed instrument categories * Flagging unclear, ambiguous, or difficult examples for review * Spotting mislabeled or low-quality files in the dataset * Keeping the dataset clean, consistent, and well organized * Working with our team to refine labelling rules where categories are unclear **What we’re looking for** * Strong ability to identify instruments by ear * Experience with music production, sound design, audio engineering, or session work * Familiarity with stem files rather than only finished mixes * Good understanding of how effects and processing can change the sound of an instrument * Careful, methodical, and consistent approach to repetitive audio work * Comfortable making judgement calls and flagging uncertainty when needed **Useful experience** * Working in a DAW * Familiarity with MIDI instruments and sample libraries * Experience with genres where instruments are heavily processed * Prior audio labelling, dataset preparation, or music tagging experience **What this role is not** This is not a model training role, a coding role, or a general data science role. We already have an ML engineer. We need someone who can help us improve the quality of the labelled audio data the model learns from. **Why this matters** The model is only as good as the labels it learns from. Accurate, consistent instrument labelling has a direct impact on how well the system performs. We’re looking for someone who enjoys careful listening, understands music production, and can help us turn messy real-world stems into a clean, reliable training dataset.