Dimension-based Speech-Quality Assessment: Development of an Estimator of the Dimension 'Noisiness'

Our aim is the development of an instrumental quality measure for telephone-band speech that is based on the assessment of three quality dimensions: "directness / frequency content", "continuity" and "noisiness". For this purpose, measures for all three quality dimensions have to be developed. This paper presents the general development procedure of these dimension estimators taking "noisiness" as an example. The implementation of such an estimator includes three basic steps: (1) the definition of parameters that describe dimension-relevant characteristics of a speech signal, (2) the development of measures that extract the dimension parameters from a speech signal, and (3) the integration of the dimension parameters into a model that provides reliable dimension estimates. Additionally, our work features the use of idealized systems. Such a system permits a systematic distortion of the parameters of one quality dimension of a speech signal. Thus, an idealized system facilitates the examination of both, the dimension parameters and the dimension model, for their appropriateness to estimate a quality dimension.