Spatially constrained vs. unconstrained filtering in neural spatiospectral filters for multichannel speech enhancement

TL;DR

This paper investigates how two different neural spatiospectral filters utilize spatial cues in multichannel speech enhancement, revealing their differing behaviors and providing insights for better design and deployment.

Contribution

It compares two neural spatiospectral filters, showing one always performs spatial filtering while the other adapts based on the task, enhancing understanding of their mechanisms.

Findings

One NSSF consistently performs spatial filtering.

The other NSSF selectively uses spatial cues depending on the task.

Insights enable informed design choices for NSSFs.

Abstract

When using artificial neural networks for multichannel speech enhancement, filtering is often achieved by estimating a complex-valued mask that is applied to all or one reference channel of the input signal. The estimation of this mask is based on the noisy multichannel signal and, hence, can exploit spatial and spectral cues simultaneously. While it has been shown that exploiting spatial and spectral cues jointly is beneficial for the speech enhancement result, the mechanics of the interplay of the two inside the neural network are still largely unknown. In this contribution, we investigate how two conceptually different neural spatiospectral filters (NSSFs) exploit spatial cues depending on the training target signal and show that, while one NSSF always performs spatial filtering, the other one is selective in leveraging spatial information depending on the task at hand. These insights provide better understanding of the information the NSSFs use to make their prediction and, thus, allow to make informed decisions regarding their design and deployment.