D3PIA: A Discrete Denoising Diffusion Model for Piano Accompaniment Generation From Lead sheet
Eunjin Choi, Hounsu Kim, Hayeon Bang, Taegyun Kwon, Juhan Nam · 2026
Read D3PIA: A Discrete Denoising Diffusion Model for Piano Accompaniment Generation From Lead sheet by Eunjin Choi, Hounsu Kim, Hayeon Bang, Taegyun Kwon, Juhan Nam (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-03523.txt.
Description
Given only a melody and its chords, a pianist can improvise a full accompaniment; teaching a model the same trick is harder. Eunjin Choi and 4 co-authors propose D3PIA in a 2026 paper, a discrete diffusion model for piano accompaniment generation from a lead sheet. The model works in piano-roll form, where notes appear as a grid over time and pitch.
D3PIA relies on local alignment between the lead sheet and the accompaniment it builds. Neighborhood Attention lets it read the lead sheet and condition each predicted note on nearby melody and chord context. On the POP909 benchmark, it preserves the given chord conditions more faithfully than continuous diffusion and transformer baselines. Listening tests also favor its accompaniments as more musically coherent.
Catalog record
| CREATOR | Eunjin Choi, Hounsu Kim, Hayeon Bang, Taegyun Kwon, Juhan Nam |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 24 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-03523