Temporal Convolution Network-based Onset Detection and Query by Humming System Design

The onset is a key factor to split an audio signal into several notes and plays a critical role in music signal processing. In this study, we ensemble multiple temporal convolution network (TCN) based models and utilize a restricted frequency range spectrogram to achieve more robust onset detection. Different from the present onset detection method in the query by humming (QBH) system which is only available in a clean scenario, the proposed onset detection algorithm prevents the noise from affecting the onset detection function (ODF). Compared to the CNN model which exploits spatial features of the spectrogram, the TCN model exploits both spatial and temporal features of the spectrogram. We apply the TCN-based enhancement as a preprocessor of the QBH architecture to address the interference of noise better. With the combinations of TCN-based speech enhancement and onset detection, the experimental results show that the proposed algorithm makes the QBH system perform accurately in both noisy and clean circumstances with less computation time.