The paper presents JEN-1 Composer, a unified AI framework for high-fidelity multi-track music generation. This system addresses the existing gap in generative AI models that lack the flexibility to compose and control separate music tracks. JEN-1 Composer can seamlessly incorporate any diffusion-based music generation system, enhancing its capacity for versatile multi-track music generation. A unique training strategy is also introduced, incrementally instructing the model from single-track to multi-track combinations. The proposed model thus represents a significant advance in interactive AI-facilitated music creation.
Publication date: 29 Oct 2023
Project Page: https://jenmusic.ai/audio-demos
Paper: https://arxiv.org/pdf/2310.19180
