Output-based Objective Speech Quality Using Vector Quantization Techniques

Output-based speech quality (OBQ) refers to an objective speech quality measure that uses only received speech without access to the input speech record. This paper proposes two new OBQ measures and evaluates their performance. Perceptual linear prediction (PLP) coefficients are used to provide speaker independence required by the objective measure. Two distortion measures are introduced for predicting speech quality based on vector quantization of the output speech record. These are the transition probability distance and the median minimum distance measure. The OBQ parameters are tested on four different speech datasets. The correlation is computed between subjective scores and the objective quality measures under a variety conditions, and the results indicate that the proposed algorithms are quite robust against speaker, text and distortion variation.