Compose & Embellish: Well-Structured Piano Performance Generation via A Two-Stage Approach
This work addresses the problem of generating well-structured piano performances for music generation applications, representing an incremental improvement over existing methods.
The paper tackles the challenge of generating expressive piano performances with long-range musical structures by proposing a two-stage Transformer-based framework that first composes a lead sheet and then embellishes it with accompaniment and expressive touches. The results show that the method reduces the gap in structureness between state-of-the-art and real performances by half and improves other musical aspects like richness and coherence.
Even with strong sequence models like Transformers, generating expressive piano performances with long-range musical structures remains challenging. Meanwhile, methods to compose well-structured melodies or lead sheets (melody + chords), i.e., simpler forms of music, gained more success. Observing the above, we devise a two-stage Transformer-based framework that Composes a lead sheet first, and then Embellishes it with accompaniment and expressive touches. Such a factorization also enables pretraining on non-piano data. Our objective and subjective experiments show that Compose & Embellish shrinks the gap in structureness between a current state of the art and real performances by half, and improves other musical aspects such as richness and coherence as well.