MusicFlow: Cascaded Flow Matching for Text Guided Music Generation
This addresses music generation from text descriptions for creative applications, offering efficiency gains but appearing incremental as it builds on existing flow matching techniques.
The authors tackled text-to-music generation by introducing MusicFlow, a cascaded flow matching model that uses self-supervised representations to bridge text and audio, resulting in music that exhibits superior quality and text coherence while being 2-5 times smaller and requiring 5 times fewer iterative steps than alternatives.
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.