Diagnostic Instrumental Speech Quality Model

Abstract

An instrumental measure should provide a correct ranking of various speech processing systems. Even though new instrumental methods have been developed for the perceptual assessment of VoIP transmissions, none of the current model correctly estimates the degradations introduced by all in-use telecommunication systems. The reliable assessment of very different connections (i.e. either a clean WB or a noisy NB stimulus) by a single universal instrumental measure is not yet possible with existing models. The current ITU-T standards show some limits in their quality estimations. They are mainly caused by the ongoing complexity of network topologies and their speech processing components. These limits have been detailed in the previous Chapter. The ITU-T launched a new standardization program called Perceptual Objective Listening Quality Analysis (POLQA) to select a new intrusive speech quality model (ITU–T Temp. Doc. TD.52 Rev.1, 2007). This selection process is detailed in Sec. 5.1.3. A new model, called Diagnostic Instrumental Assessment of Listening quality (DIAL), developed as part of this standardization program is presented in this chapter. This model relies on a specific framework: it combines a Core model, based on TOSQA, and four dimension estimators. The Core model assesses the nonlinear degradations introduced by a speech transmission system, whereas the dimension estimators quantify the linear degradations on the four perceptual dimensions defined in Sec. 1.4: Coloration, Loudness, Discontinuity and Noisiness. Then, an aggregation of all the degradations simulates a cognitive process employed by human subject during the quality judgment process. In accordance with the development procedure of an instrumental model introduced in Sec. 2.3, this chapter presents the second step, i.e. results of auditory tests are already available and a candidate measure has been developed using those results. The validation process of this candidate model is presented in Chap. 5.