MetricNet: Towards Improved Modeling For Non-Intrusive Speech Quality Assessment

TL;DR

This paper introduces MetricNet, a novel non-intrusive speech quality assessment model that leverages label distribution and speech reconstruction learning to achieve high correlation with intrusive evaluation methods.

Contribution

The paper presents a new non-intrusive speech quality measurement model, MetricNet, which significantly improves performance over existing models using innovative learning techniques.

Findings

High correlation with intrusive speech quality evaluation.

Effective on clean, noisy, and processed speech data.

Outperforms existing non-intrusive models.

Abstract

The objective speech quality assessment is usually conducted by comparing received speech signal with its clean reference, while human beings are capable of evaluating the speech quality without any reference, such as in the mean opinion score (MOS) tests. Non-intrusive speech quality assessment has attracted much attention recently due to the lack of access to clean reference signals for objective evaluations in real scenarios. In this paper, we propose a novel non-intrusive speech quality measurement model, MetricNet, which leverages label distribution learning and joint speech reconstruction learning to achieve significantly improved performance compared to the existing non-intrusive speech quality measurement models. We demonstrate that the proposed approach yields promisingly high correlation to the intrusive objective evaluation of speech quality on clean, noisy and processed speech data.