Rhythmic Stability and Synchronization in Multi-Track Music Generation

Current paradigms for multi-track music generation and assessment notably overlook two perceptually crucial elements: beat stability and cross-track synchronization, despite their demonstrated impact on listener experience. This paper presented three novel evaluation metrics for multi-track music generation, establishing a comprehensive evaluation framework for multi-track music quality compared with existing standards. Moreover, we proposed a synchronous multi-track waveform generation model, SyncTrack. Experiments demonstrate that SyncTrack achieves superior performance on both conventional and newly proposed metrics, validating the effectiveness of our model and the utility of our evaluation framework.