Skip to content
Haoran Hou
← ResearchPersonal music assistant · research proposal
Personal music assistant · research proposal

ohmate

A music assistant that learns from a creator’s decisions, with room to ask, correct, and change your mind.

ohmate
Interface concept: audition a shorter fill alongside the working version, then choose whether to keep the change.

What should an assistant remember?

A producer may choose a shorter drum fill because a vocal is still going, because this particular song needs space, or because they simply prefer it today. Those choices can look identical in an interaction log. ohmate asks how an assistant could learn from musical work while keeping its interpretation open to correction.

Keep the working version intact

In the proposed workflow, A is the working project and B is an alternative prepared within a scope the creator has agreed to. The creator can compare them, keep selected changes, ask for another direction, or stop. Only the creator applies changes to A. Questions that help with the current musical task can happen during work; optional questions for learning about the creator wait until the task is finished.

Events, interpretations, and corrections

The design separates what happened, what the assistant thinks it means, and what the creator actually said. An undo is recorded without assuming a reason. A correction such as “leave space under this lead part” can apply to that musical situation without becoming a general preference. The record keeps the original choices even when an interpretation is withdrawn.

The paper and the next test

The public preprint includes a reproducible analytical example following compact and fuller fill choices across projects, an unexplained undo, and a scoped correction. It checks the behaviour of the proposed update rules. The integrated assistant has not been evaluated with users. The proposed evaluation would compare recommendations with and without personal history, then examine whether creators can find and correct mistaken interpretations. The conceptual paper is submitted to EAIM 2027.

Read PDF: ohmate as a personal music assistant: What to learn from a creator’s choices ↗

Haoran Hou · independent research and system design