Supervised Metric Learning for Music Structure Features

Abstract

Music structure analysis (MSA) methods traditionally search for musically meaningful patterns in audio: homogeneity, repetition, novelty, and segment-length regularity. Hand-crafted audio features such as MFCCs or chromagrams are often used to elicit these patterns. However, with more annotations of section labels (e.g., verse, chorus, bridge) becoming available, one can use supervised feature learning to make these patterns even clearer and improve MSA performance. To this end, we take a supervised metric learning approach: we train a deep neural network to output embeddings that are near each other for two spectrogram inputs if both have the same section type (according to an annotation), and otherwise far apart. We propose a batch sampling scheme to ensure the labels in a training pair are interpreted meaningfully. The trained model extracts features that can be used in existing MSA algorithms. In evaluations with three datasets (HarmonixSet, SALAMI, and RWC), we demonstrate that using the proposed features can improve a traditional MSA algorithm significantly in both intra-and cross-dataset scenarios.

Barwise Compression Schemes for Audio-Based Music Structure Analysis

“…Segmentation is usually based on criteria such as homogeneity, novelty, repetition and regularity [1]. When performed algorithmically, MSA often relies on similarity criteria within passages of a song summarized in an autosimilarity matrix [2][3][4][5][6][7][8], in which each coefficient represents an estimation of the similarity between two musical fragments.…”

“…While similarity between two frames can be obtained from the feature representation of the signal, such as the STFT of the song [2], recent works try to design new representations of the original music, able to capture the similarity between two frames while maintaining a high level of dissimilarity between dissimilar frames [1,[3][4][5][6][7][8]. This generally consists in projecting the data in a new feature space and computing the similarity in the feature space.…”

“…Instead, the present work considers that repetitions are more prone to happen at the barscale, and hence focuses on barwise aligned features, as in [7]. A comparison between both has been made in [8], without one singling out.…”