Assessing the Impact of Noise and Speech Enhancement on the Intelligibility of Speech Codecs

TL;DR

This paper evaluates how noise and speech enhancement affect the intelligibility of classical and neural speech codecs, highlighting the robustness of classical codecs and the benefits of speech enhancement in noisy conditions.

Contribution

It provides a comparative analysis of classical and neural codecs under noisy conditions and assesses the impact of speech enhancement on intelligibility and listening effort.

Findings

Classical codecs are more noise robust than neural codecs.

Speech enhancement improves intelligibility and reduces listening effort in noisy scenarios.

Objective intelligibility measures correlate highly with subjective scores.

Abstract

Preserving speech intelligibility is a minimum requirement for speech codecs in communication. Recently, very low-bitrate neural codecs have gained interest for replacing classical codecs, reinforcing the need to evaluate whether intelligibility is preserved in realistic scenarios. In this paper, we evaluate the intelligibility and listening effort of classical and neural speech codecs in clean and noisy conditions. Further, we assess the impact of speech enhancement (SE) before coding, simulating a possible audio processing pipeline. The results show that classical codecs are more noise robust than neural codecs. Further, SE can lead to significant intelligibility and listening effort improvements for codecs otherwise negatively affected by noise. Listening effort reveals nuanced differences when intelligibility is saturated. Lastly, objective intelligibility based on automatic speech recognition is highly correlated with subjective intelligibility scores averaged per condition.