Wave

What we build

Wave is a standing team of audio-industry veterans with decades in studios and control rooms. We produce custom, expert-annotated audio datasets for AI teams, obsessive about labeling, filtering, and curation.

Supervised labels lead: tasks, traces, captions, attributes, corrections, and segment-level annotation, built to a spec we co-design with you. Every engagement runs the same way — spec co-design, a small pilot set you can validate, then the full build.

Why the ears matter

Expert human judgment is the ceiling automated tools are still chasing: general AI judges reach only 60–70% agreement with expert listeners, and objective metrics correlate weakly with what people actually hear. A dataset is only as good as the people who label it. Read our analysis: Can AI judge music quality? What the data says in 2026.

FAQ

What does Wave do?

Wave is a team of audio-industry veterans producing custom, expert-annotated audio datasets for AI teams: tasks, traces, captions, attributes, corrections, and segment-level labels, built to a co-designed spec.

Who annotates the audio?

Working audio professionals with decades in studios and control rooms — not an anonymous crowd. A dataset is only as good as the ears that label it.

How does an engagement work?

We co-design the labeling spec with your team, deliver a small pilot set you can validate, then scale to the full build.

Why does annotation quality matter for audio models?

Because expert human judgment is the standard automated tools chase: general AI judges reach only 60–70% agreement with expert listeners. Models learn from the labels they are given.