Robust Multipitch Estimation of Piano Sounds Using Deep Spiking Neural Networks

In this paper, we propose a robust multi-label classification system based on deep spiking neural networks to handle multi-pitch estimation tasks. We employ constantQ transform spectrogram as a time-frequency representation. A keypoint detection technique is used for noise suppression and the extraction of relevant information. We also propose a novel biological spiking coding method that fits the expression of musical signals. This coding method can encode time, frequency, intensity information into spatiotemporal spike trains. And the spatio-temporal credit assignment (STCA) algorithm is used to train deep spiking neural networks. We perform the multipitch evaluation on the MAPS data set, and our work compares with the state-of-the-art methods by using the F1-score metric. Experimental results show that the proposed scheme has achieved better performance than other state-of-the-art methods and reveal the system's robustness to environmental noise.