Instabilities in Convnets for Raw Audio

TL;DR

This paper investigates why training convolutional neural networks for raw audio is challenging, focusing on the impact of initialization and filter size on stability and approximation quality.

Contribution

It introduces a theoretical framework analyzing large deviations in filterbank responses, highlighting issues with large filters and periodic signals in audio convnets.

Findings

Deviations increase with larger filters and periodic inputs.

Numerical simulations confirm the theory.

Condition number scales logarithmically with filter size.

Abstract

What makes waveform-based deep learning so hard? Despite numerous attempts at training convolutional neural networks (convnets) for filterbank design, they often fail to outperform hand-crafted baselines. These baselines are linear time-invariant systems: as such, they can be approximated by convnets with wide receptive fields. Yet, in practice, gradient-based optimization leads to suboptimal approximations. In our article, we approach this phenomenon from the perspective of initialization. We present a theory of large deviations for the energy response of FIR filterbanks with random Gaussian weights. We find that deviations worsen for large filters and locally periodic input signals, which are both typical for audio signal processing applications. Numerical simulations align with our theory and suggest that the condition number of a convolutional layer follows a logarithmic scaling law between the number and length of the filters, which is reminiscent of discrete wavelet bases.