MusicFlow: Cascaded Flow Matching for Text Guided Music Generation

TL;DR

MusicFlow is a novel cascaded flow matching model for text-guided music generation that produces high-quality, coherent music efficiently and can generalize to related tasks like music infilling and continuation.

Contribution

The paper introduces MusicFlow, a flow matching-based model that effectively bridges text and music, enabling zero-shot infilling and continuation with fewer resources.

Findings

MusicFlow generates music with superior quality and coherence.

The model is smaller and faster than previous methods.

It achieves competitive results in music infilling and continuation.

Abstract

We introduce MusicFlow, a cascaded text-to-music generation model based on flow matching. Based on self-supervised representations to bridge between text descriptions and music audios, we construct two flow matching networks to model the conditional distribution of semantic and acoustic features. Additionally, we leverage masked prediction as the training objective, enabling the model to generalize to other tasks such as music infilling and continuation in a zero-shot manner. Experiments on MusicCaps reveal that the music generated by MusicFlow exhibits superior quality and text coherence despite being over $2\sim5$ times smaller and requiring $5$ times fewer iterative steps. Simultaneously, the model can perform other music generation tasks and achieves competitive performance in music infilling and continuation. Our code and model will be publicly available.