When developing deep learning models for melody extraction, MedleyDB is a must-have dataset owing to its scale, diverse genres, and quality. However, it has no ground-truth labels for vocal melody. At the same time, none of existing methods for vocal melody labeling is able to accurately label vocal melody. In this paper we propose a simple, correct method for this task. Through extensive experiment, we evaluate the influence of labeling on the performance of models for vocal melody extraction, and find that in most cases the proposed method can boost the performance. The proposed method can accelerate research in relevant areas.

Melody extraction is a challenging task in music information retrieval that enables many down-stream applications. In this paper we propose a simple dilated convolutional model for melody extraction. It takes variable-q transforms as inputs. It first uses consecutive layers of convolution to capture local temporal-frequency patterns. Afterward, it relies only a single layer of dilated convolution for capturing global frequency patterns formed by the pitches and harmonics of active notes. This model is effective in that it achieves the-state-of-the-art performance on most datasets, for both general and vocal melody extraction. In addition, it gets the best performance with the least training data.