Optimal Embedding Calibration for Symbolic Music Similarity

TL;DR

This paper introduces an embedding calibration method for music similarity that leverages composer information to automatically evaluate and improve the performance of pre-trained models, surpassing baseline methods.

Contribution

It proposes a novel calibration approach using composer data to optimize music embedding similarity, filling a gap in automatic evaluation without human labels.

Findings

Optimal calibration method outperforms baselines

Composer-based labels enable automatic evaluation

Significant improvement in music similarity metrics

Abstract

In natural language processing (NLP), the semantic similarity task requires large-scale, high-quality human-annotated labels for fine-tuning or evaluation. By contrast, in cases of music similarity, such labels are expensive to collect and largely dependent on the annotator's artistic preferences. Recent research has demonstrated that embedding calibration technique can greatly increase semantic similarity performance of the pre-trained language model without fine-tuning. However, it is yet unknown which calibration method is the best and how much performance improvement can be achieved. To address these issues, we propose using composer information to construct labels for automatically evaluating music similarity. Under this paradigm, we discover the optimal combination of embedding calibration which achieves superior metrics than the baseline methods.