Sonautic

I'm building an ML piano performance evaluation system for music education, not just "did you hit the right notes," but actual musical assessment: tone quality, phrasing, dynamics, interpretation, musicality. The kind of feedback conservatory teachers give, but scalable to 225M+ musicians worldwide. My outreach goal: Get feasibility validation and technical advice from audio ML researchers whose work I'm building on, specifically the MERT authors (the music-specific transformer I'm using) and other experts in music information retrieval, multi-modal learning, and audio ML. I'm a Berklee classical pianist who pivoted to ML engineering, and I needed to know if my architecture is feasible with limited expert labels (~300), or if I'm being naively optimistic. The stakes: I've spent 2+ weeks designing this system, translating conservatory evaluation rubrics into ML objectives. If experts say "this won't work," I need to know before investing in expensive data annotation.