Learning Jazz Pianist Style with Cross-Attention Conditioning

Drew Edwards; Akira Maezawa; Simon Dixon

cs.SDOct 2, 2026

Learning Jazz Pianist Style with Cross-Attention Conditioning

Authors: Drew Edwards, Akira Maezawa, Simon Dixon

Organizations: Centre for Digital Music, Queen Mary University of London · Yamaha Corporation Hamamatsu, Japan

Abstract

Jazz pianists develop distinctive traits that experienced listeners can often identify within seconds, yet the features underlying this recognition resist formal description. We study jazz pianist style through the lens of a pretrained symbolic music transformer, showing that its learned representations already encode pianist identity well enough for highly accurate classification across two benchmarks. We then augment the transformer with cross-attention over learned pianist identity embeddings, enabling it to generate music conditioned on a specific artist's style. Two evaluation protocols confirm that the generator captures meaningful stylistic structure: a sliding-window classifier consistently attributes conditioned continuations to the correct artist, far above unconditioned baselines; and a classifier trained entirely on synthetic generations identifies real pianists across 12 classes with 87% chunk-level and 95% song-level accuracy. Finally, we repurpose the classifier to locate the most characteristic moments within a performance, surfacing the specific musical gestures that distinguish each pianist's voice.

Figures & tables

Figure 1: t-SNE [ 20 ] projection of Aria-CL track embeddings for all PiJAMA-30 pianists. The 12 selected pianists (colored) occupy distinct clusters; remaining artists are shown in light gray.

Method / Dataset	Chunk	Clip	Track
ResNet-50 [ 5 ] on DPI-20	—	80.5	94.4
Ours (Aria) on DPI-20	84.1	90.4	96.9
Ours on PiJAMA-12	95.8	—	98.8 †
Ours on PiJAMA-30	90.2	—	99.4 †

Table 1: Pianist identification accuracy (%). A chunk is a 1024-token classification window; a clip is a

\sim

30 s audio segment following the protocol of [ 5 ] . Top: Deep Pianist benchmark (20 classes) using published splits; clip and track columns aggregate windows by mean logit. Bottom: PiJAMA classifiers (12 and 30 classes, identical recipe) on held-out song-level test splits (80 and 156 songs); the 12-class model is used for all style-fidelity evaluations. † Majority vote, ties broken by mean logit; each classifier misses one short track (of 80 and 156).

Figure 2: Gated cross-attention block added to the last 8 of 16 transformer layers. Queries come from the music sequence; keys and values from learned artist embeddings.

Model	Test NLL	Perplexity	Agreement †
Pretrained Aria	2.43	11.41	25%
Fine-tuned (no CA)	1.94	6.96	37%
Cross-attention (ours)	1.92	6.82	70%

Table 2: Token-weighted cross-entropy on the PiJAMA-12 test split (177 sequences, 562 125 tokens) and mean classifier agreement during free generation. Despite near-identical perplexities, style fidelity differs dramatically. † Mean sliding-window agreement at

P{=}256

(Figure 3 ).

Figure 3: Classifier agreement with the true artist as a function of window-start position along a 4096-token continuation. Three panels show prompt lengths

P\in\{128,256,512\}

. The conditioned model (blue) sustains a

\sim

33-point advantage over the fine-tuned baseline across the full generation. The ablation (dashed red) tracks the fine-tuned baseline, isolating the effect of the conditioning signal from the architecture.

Figure 4: Per-artist conditioned agreement at

P{=}256

(Section 5 ). Sample counts shown at right. Agreement varies from 96% (Hank Jones) to 29% (Cedar Walton); the dotted line marks chance (8.3%).

Figure 5: Per-pianist chunk-level test accuracy on PiJAMA-12. Blue: real-data classifier (Section 4.1 ). Orange: synthetic-only classifier (Section 6 ). Artists are ordered by synthetic accuracy.

Training data	Chunk	Song †
Real PiJAMA-12	95.8	98.8
+ from scratch (ResNet-50)	93.1 ‡	97.5
Synthetic only (ours)	87.1	95.0
+ from scratch (ResNet-50)	70.2 ‡	91.3

Table 3: Pianist identification on the real PiJAMA-12 test set (80 songs, 591 chunks). Synthetic-trained classifiers see only generated continuations; within each architecture, the real- and synthetic-trained classifiers share one training recipe. † Majority vote over chunks within a song; exact vote ties are broken by mean logit. ‡ 30 s piano-roll clips rather than token chunks.

Figure 6: Characteristic region detection for Art Tatum, “Sophisticated Lady.” Top: within-track z-score of the classifier’s logit margin; red fill marks confidence above the track average, blue below. Bottom: piano rolls of the most (left, $z{=}1.9$ ) and least (right, $z{=}{-}1.2$ ) characteristic 15-second regions, notes colored by z-score. The peak’s dense runs span over five octaves, consistent with Tatum’s virtuosic runs; the trough shows comparable note density (9.7 vs. 10.6 notes/s) over a narrower range (4.4 vs. 5.2 octaves), suggesting the classifier’s confidence tracks specific gestures, not overall activity.

Music Generation

Music generation studies computational methods for composing, synthesizing, and editing music from symbolic, audio, textual, and multimodal inputs.

See all 107 papers

Learning Music Style for Piano Arrangement Through Cross-Modal Bootstrapping
Aug 4, 2026Jingwei Zhao, Gus Xia, Ziyu Wang +1Controllable Music Generation Cross-Modal Alignment
1Songscription · School of Computing, National University of Singapore · Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) +1
Pianist Transformer: Towards Expressive Piano Performance Rendering via Scalable Self-Supervised Pre-Training
Dec 2, 2025Hong-Jie You, Jie-Jing Shao, Xiao-Wen Yang +3Efficient Transformer Inference Self-Supervised Pre-Training
State Key Laboratory of Novel Software Technology, Nanjing University, Nanjing 210023, China · School of Artificial Intelligence, Nanjing University, Nanjing 210023, China · School of Intelligence Science and Technology, Nanjing University, Suzhou 215163, China
Structure-Aware Piano Accompaniment via Style Planning and Dataset-Aligned Pattern Retrieval
Feb 16, 2026Wanyu Zang, Yang Yu, Meng YuMusic Accompaniment Generation Music Generation
Computer Science, Lewis University · Nexus Art Foundation · Computer Science, Governors State University

Aug 4, 2026J/K move · Enter open · S save

Learning Music Style for Piano Arrangement Through Cross-Modal Bootstrapping

Jingwei Zhao, Gus Xia, Ziyu Wang, Ye Wang

1Songscription · School of Computing, National University of Singapore · Mohamed bin Zayed University of Artificial Intelligence (MBZUAI)+1

Controllable Music Generation Cross-Modal Alignment Symbolic Music Generation Music Generation

Open paper AskPDF arXiv

What is music style? Though often described using text labels such as "swing," "classical," or "emotional," the real style remains implicit and hidden in concrete music examples. In this paper, we introduce a cross-modal framework that learns implicit music styles from raw audio and applies them to symbolic music generation. Inspired by BLIP-2, our model leverages a Querying Transformer (Q-Former) to extract style representations from a large, pre-trained audio language model (LM), and further applies them to condition a symbolic LM for generating piano arrangements. We adopt a two-stage training strategy: contrastive learning to align auditory style with symbolic expression, followed by generative modeling for music arrangement. Our model generates piano performances jointly conditioned on a lead sheet (content) and a reference audio example (style), enabling controllable and stylistically faithful arrangement. Experiments demonstrate the effectiveness of our approach in piano cover generation, style transfer, and audio-to-MIDI retrieval, achieving substantial improvements in style-aware alignment and music quality.

Abstract

Music Generation

Explore similar work

Learning Music Style for Piano Arrangement Through Cross-Modal Bootstrapping

Pianist Transformer: Towards Expressive Piano Performance Rendering via Scalable Self-Supervised Pre-Training

Structure-Aware Piano Accompaniment via Style Planning and Dataset-Aligned Pattern Retrieval