Subjective Evaluation of Deep Neural Network Based Speech Enhancement Systems in Real-World Conditions

TL;DR

This study compares deep neural network-based speech enhancement systems to traditional Wiener-filter methods in real-world conditions, showing DNNs improve noise suppression with minimal impact on speech quality and intelligibility.

Contribution

It provides a subjective evaluation of DNN-based speech enhancement in real-world scenarios, highlighting their advantages over traditional methods.

Findings

DNNs outperform Wiener-filter in noise suppression across conditions.

DNNs maintain speech quality and intelligibility better than traditional methods.

DNNs do not significantly degrade speech quality or noise transparency.

Abstract

Subjective evaluation results for two low-latency deep neural networks (DNN) are compared to a matured version of a traditional Wiener-filter based noise suppressor. The target use-case is real-world single-channel speech enhancement applications, e.g., communications. Real-world recordings consisting of additive stationary and non-stationary noise types are included. The evaluation is divided into four outcomes: speech quality, noise transparency, speech intelligibility or listening effort, and noise level w.r.t. speech. It is shown that DNNs improve noise suppression in all conditions in comparison to the traditional Wiener-filter baseline without major degradation in speech quality and noise transparency while maintaining speech intelligibility better than the baseline.