MAJEPPA: Morphing and Assessing in a Unified Piano Performance Space

2026-08-11

AI Summary

This work proposes the first unified self-supervised representation framework for modeling piano performance across skill levels, from beginners to virtuosos. Addressing the challenge of integrating performance generation and evaluation across diverse proficiency tiers, the method jointly learns representations of musical scores and expressive performances within a shared embedding space by combining next-token prediction, InfoNCE loss, and supervised contrastive loss. The authors construct a large-scale piano performance dataset encompassing six skill levels and six recording conditions, leveraging a pretrained MIDI autoregressive model for conditional generation and multidimensional assessment. Evaluated on the newly introduced EVPMR benchmark, the model significantly outperforms existing approaches in tasks including performance quality assessment, competition ranking prediction, error detection, and technical skill classification, demonstrating the effectiveness and generalizability of its learned representations.

Abstract

We present MAJEPPA, a self-supervised framework to learn piano performance representations that span the full skill spectrum, from beginner practice sessions to virtuoso concert recordings. We curate the MAJEPPA dataset, comprising ~4,000 annotated recordings across six expertise levels and six recording contexts. We adapt a single pre-trained MIDI autoregressive model with a joint objective: next-token prediction learns score-conditioned performance generation at various skill levels, while InfoNCE and supervised contrastive losses align abstract score and performance representations in a joint embedding space. The proposed model both generates and understands performances in a unified framework. By introducing the EVPMR benchmark, a suite of downstream tasks spanning quality assessment, competition ranking, mistake and technique classification, we evaluate the learnt representations, demonstrating progress towards a real-world model for the piano performance space.

Problem

• piano performance
• skill spectrum
• performance representation
• quality assessment
• mistake classification

Innovation

• self-supervised learning
• piano performance representation
• joint embedding space
• contrastive learning
• generative modeling