AECMOS: A speech quality assessment metric for echo impairment
This provides a more efficient and correlated tool for evaluating echo cancellation models, addressing a domain-specific need in speech processing.
The paper tackled the problem of evaluating acoustic echo cancellers by developing AECMOS, a neural network model that assesses speech quality degradation due to echo and other sources, showing accuracy through correlation with human subjective ratings.
Traditionally, the quality of acoustic echo cancellers is evaluated using intrusive speech quality assessment measures such as ERLE \cite{g168} and PESQ \cite{p862}, or by carrying out subjective laboratory tests. Unfortunately, the former are not well correlated with human subjective measures, while the latter are time and resource consuming to carry out. We provide a new tool for speech quality assessment for echo impairment which can be used to evaluate the performance of acoustic echo cancellers. More precisely, we develop a neural network model to evaluate call quality degradations in two separate categories: echo and degradations from other sources. We show that our model is accurate as measured by correlation with human subjective quality ratings. Our tool can be used effectively to stack rank echo cancellation models. AECMOS is being made publicly available as an Azure service.