Skip to content
Conference

A learning method for multitask recognition of piano playing dynamics and articulation

Oct 2026 · The International Conference on Optical Imaging and Deep Learning · Vol 14494, pp. 1449432 - 1449432-9 · 0 citations · 15 references
Engineering

Abstract

This paper presents a multitask learning framework for the simultaneous recognition of piano playing dynamics and articulation from audio recordings. The proposed system integrates advanced audio feature extraction methods, including mel-frequency cepstral coefficients and constant-Q transforms, with precise onset and offset detection to capture both spectral and temporal information relevant to expressive performance. A hierarchical Transformer neural network with modality-adaptive attention fusion is developed. The neural network is used to model, disentangle shared and task-specific representations. The recognition accuracy and robustness of the framework are significantly improved compared with the established baselines based on the widely annotated dataset. The hierarchical specialization and adaptive attention modules contribute significantly to the system performance. According to detailed case studies, the attention mechanism always focuses on musically significant events, thus providing interpretable and reliable predictions. This study establishes a set of technically rigorous methods for automatically analyzing performance parameters of piano music. These methods support applications such as intelligent performance evaluation and computational music analysis.

View source

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.