Basic Filters for Convolutional Neural Networks Applied to Music: Training or Design?

TL;DR

This paper investigates whether applying adaptive or learned filters directly to raw music data can outperform traditional spectrogram-based features in CNNs, showing that adaptive filters can improve singing voice detection.

Contribution

It provides a theoretical and experimental comparison between traditional spectrogram features and adaptive filter banks for CNN-based music classification.

Findings

Adaptive filter banks can approximate mel-spectrograms.

Adaptive features outperform traditional spectrograms in singing voice detection.

Learned filter parameters perform as well as fixed adaptive filters.

Abstract

When convolutional neural networks are used to tackle learning problems based on music or, more generally, time series data, raw one-dimensional data are commonly pre-processed to obtain spectrogram or mel-spectrogram coefficients, which are then used as input to the actual neural network. In this contribution, we investigate, both theoretically and experimentally, the influence of this pre-processing step on the network's performance and pose the question, whether replacing it by applying adaptive or learned filters directly to the raw data, can improve learning success. The theoretical results show that approximately reproducing mel-spectrogram coefficients by applying adaptive filters and subsequent time-averaging is in principle possible. We also conducted extensive experimental work on the task of singing voice detection in music. The results of these experiments show that for classification based on Convolutional Neural Networks the features obtained from adaptive filter banks followed by time-averaging perform better than the canonical Fourier-transform-based mel-spectrogram coefficients. Alternative adaptive approaches with center frequencies or time-averaging lengths learned from training data perform equally well.