Towards Explaining Expressive Qualities in Piano Recordings: Transfer of Explanatory Features Via Acoustic Domain Adaptation

Abstract

Emotion and expressivity in music have been topics of considerable interest in the field of music information retrieval. In recent years, mid-level perceptual features have been suggested as means to explain computational predictions of musical emotion. We find that the diversity of musical styles and genres in the available dataset for learning these features is not sufficient for models to generalise well to specialised acoustic domains such as solo piano music. In this work, we show that by utilising unsupervised domain adaptation together with receptive-field regularised deep neural networks, it is possible to significantly improve generalisation to this domain. Additionally, we demonstrate that our domain-adapted models can better predict and explain expressive qualities in classical piano performances, as perceived and described by human listeners.

Understanding Feature Importance in Musical Works: Unpacking Predictive Contributions to Cluster Analyses

“…Researchers increasingly apply machine learning techniques to clarify relationships between music's structural features (from low-level acoustic features to mid-level perceptual ones) and participants’ perceptual responses (Chowdhury & Widmer, 2021; Cowen et al, 2020; Zhukov, 2007). Estimating the importance of analyzed features can help researchers identify which ones to assess in perceptual experiments.…”

“…We analyzed a data set containing the number of distinct note attacks, pitch height, and nominal mode (as defined by the composer) from the first eight measures (appending pick-up measures where applicable) of each prelude in Bach's The Well-Tempered Clavier and Chopin's Preludes . Previous studies in empirical musicology (Poon & Schutz, 2015) and music psychology (e.g., Anderson & Schutz, 2022; Battcock & Schutz, 2022; Chowdhury & Widmer, 2021) have analyzed these corpora, providing an opportunity for exploring the broader musicological and perceptual implications of these analyses. Here, we limited our selection to eight-measure excerpts (appending pick-up measures where applicable) for methodological consistency with parallel work involving perceptual experiments.…”

“…Previous studies in empirical musicology (Poon & Schutz, 2015) and music psychology (e.g., Anderson & Schutz, 2022;Battcock & Schutz, 2022;Chowdhury & Widmer, 2021) have analyzed these corpora, providing an opportunity for exploring the broader musicological and perceptual implications of these analyses. Here, we limited our selection to eight-measure excerpts (appending pick-up measures where applicable) for methodological consistency with parallel work involving perceptual experiments.…”

On Perceived Emotion in Expressive Piano Performance: Further Experimental Evidence for the Relevance of Mid-level Perceptual Features

“…We learn these features from the Mid-level Dataset [15] using a receptive-field regularised residual neural network (RF-ResNet) model [17]. Since we intend to use this model to extract features from solo piano recordings (a genre that is not covered by the original training data), we use a domainadaptive training approach as described in [18]. We use an input audio length of 30 seconds, padded or cropped as required.…”

Expressivity-aware Music Performance Retrieval using Mid-level Perceptual Features and Emotion Word Embeddings

“…In particular, we focus on so-called Mid-level Perceptual Features -relatively high-level musical qualities that are considered to be perceptually important [5]; in our work, these are learnt from human annotations. Previous research has shown these features to have the capacity to predict musical emotions as well as to disentangle different performances based on emotion [7][8][9].…”