Search papers, labs, and topics across Lattice.
The paper introduces a structure-aware symbolic piano accompaniment system that separates high-level style planning from note-level realization. They use a transformer to predict a per-measure style plan based on musical structure and harmony, and then retrieve and reharmonize piano patterns from a corpus using an energy-based pattern matching approach. Experiments demonstrate that this approach generates diverse and stylistically consistent long-form accompaniments, with ablations quantifying the importance of the style planner.
Achieve diverse and stylistically consistent long-form piano accompaniments by explicitly planning style at the measure level and retrieving suitable patterns from a corpus.
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.
DAMO