Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

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

Wanyu Zang, Yang Yu, Meng Yu

TL;DR

A structure-aware approach for symbolic piano accompaniment that decouples high-level planning from note-level realization is introduced and style-planner-guided retrieval produces diverse long-form accompaniments with strong style realization.

Abstract

We introduce a structure-aware approach for symbolic piano accompaniment that decouples high-level planning from note-level realization. A lightweight transformer predicts an interpretable, per-measure style plan conditioned on section/phrase structure and functional harmony, and a retriever then selects and reharmonizes human-performed piano patterns from a corpus. We formulate retrieval as pattern matching under an explicit energy with terms for harmonic feasibility, structural-role compatibility, voice-leading continuity, style preferences, and repetition control. Given a structured lead sheet and optional keyword prompts, the system generates piano-accompaniment MIDI. In our experiments, transformer style-planner-guided retrieval produces diverse long-form accompaniments with strong style realization. We further analyze planner ablations and quantify inter-style isolation. Experimental results demonstrate the effectiveness of our inference-time approach for piano accompaniment generation.

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

TL;DR

A structure-aware approach for symbolic piano accompaniment that decouples high-level planning from note-level realization is introduced and style-planner-guided retrieval produces diverse long-form accompaniments with strong style realization.

Abstract

We introduce a structure-aware approach for symbolic piano accompaniment that decouples high-level planning from note-level realization. A lightweight transformer predicts an interpretable, per-measure style plan conditioned on section/phrase structure and functional harmony, and a retriever then selects and reharmonizes human-performed piano patterns from a corpus. We formulate retrieval as pattern matching under an explicit energy with terms for harmonic feasibility, structural-role compatibility, voice-leading continuity, style preferences, and repetition control. Given a structured lead sheet and optional keyword prompts, the system generates piano-accompaniment MIDI. In our experiments, transformer style-planner-guided retrieval produces diverse long-form accompaniments with strong style realization. We further analyze planner ablations and quantify inter-style isolation. Experimental results demonstrate the effectiveness of our inference-time approach for piano accompaniment generation.
Paper Structure (42 sections, 11 equations, 2 figures, 5 tables)

This paper contains 42 sections, 11 equations, 2 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Why we do not benchmark against note-level generators.
  4. Design Overview
  5. Architecture
  6. Dataset and Representation
  7. Transformer Style Planner
  8. Problem Formulation.
  9. Input Representation (Conditioning Tokens + Structure Vector).
  10. Transformer Design and Rationale.
  11. Measure-Level Pooling and Multi-Head Slot Prediction.
  12. Training Objective (Masked Style Slots).
  13. Loss Function.
  14. Supporting Style Preferences and Keyword Prompts.
  15. Inventory-Constrained Decoding as Projection.
  16. ...and 27 more sections

Figures (2)

  • Figure 1: Overview of our style-planning and retrieval-based accompaniment pipeline. Offline preparation derives discrete measure-level style labels (dynamics, articulation, rhythm, tension, texture, register) and builds a measure-level index from POP909. Online generation predicts per-measure style slots with a Transformer style planner, then retrieves and adapts dataset patterns to the target harmony with voice-leading-aware selection. Solid arrows indicate runtime flow; dashed arrows indicate offline assets.
  • Figure 2: Training dynamics of the Transformer style planner (20 epochs). The model converges quickly, with validation loss and accuracy peaking early.