This paper presents a new approach for music style transfer using diffusion models. This approach effectively captures musical attributes with minimal data and introduces a novel time-varying textual inversion module to capture mel-spectrogram features at different levels. The method can transfer the style of specific instruments and incorporate natural sounds to compose melodies. The study also proposes a bias-reduced stylization technique to obtain stable results during inference.
Publication date: 23 Feb 2024
Project Page: https://lsfhuihuiff.github.io/MusicTI/
Paper: https://arxiv.org/pdf/2402.13763
