Discovering and Steering Interpretable Concepts in Large Generative Music Models

TL;DR

This paper introduces a scalable method using sparse autoencoders to discover and steer interpretable concepts in large autoregressive music models, revealing both known and novel musical patterns.

Contribution

It presents a novel approach for extracting interpretable features from transformer-based music models and demonstrates how these concepts can be used to control model outputs.

Findings

Revealed both traditional and novel musical concepts in models

Developed scalable automated labeling and validation pipelines

Showed concepts can be used to steer model generations

Abstract

The fidelity with which neural networks can now generate content such as music presents a scientific opportunity: these systems appear to have learned implicit theories of such content's structure through statistical learning alone. This offers a potentially new lens on theories of human-generated media. When internal representations align with traditional constructs (e.g. chord progressions in music), they show how such categories can emerge from statistical regularities; when they diverge, they expose limits of existing frameworks and patterns we may have overlooked but that nonetheless carry explanatory power. In this paper, focusing on autoregressive music generators, we introduce a method for discovering interpretable concepts using sparse autoencoders (SAEs), extracting interpretable features from the residual stream of a transformer model. We make this approach scalable and evaluable using automated labeling and validation pipelines. Our results reveal both familiar musical concepts and coherent but uncodified patterns lacking clear counterparts in theory or language. As an extension, we show such concepts can be used to steer model generations. Beyond improving model transparency, our work provides an empirical tool for uncovering organizing principles that have eluded traditional methods of analysis and synthesis.