AECMOS: A speech quality assessment metric for echo impairment

  • Marju Purin ,
  • Sten Sootla ,
  • Mateja Sponza ,
  • Ando Saabas ,

ICASSP 2022 |

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.

Publication Downloads

AEC-MOS

May 25, 2022

We have created AECMOS for evaluating clips with regards to echo ratings and other degradations ratings. There are two ways for you to use AECMOS: a web API and an onnx version of the AECMOS model. We recommend the second way, using the onnx model, because it can be much faster than the web API. The onnx model and inference script are located in the AECMOS_local directory.