Disordered Speech Quality estimation using the Matching Pursuit algorithm

This paper proposes a novel non-intrusive auditory perception-based approach for disordered speech quality estimation. An adaptive time-frequency algorithm, viz. the Matching Pursuit (MP) algorithm, is used to generate a reference signal from the disordered speech signal. Both the generated reference signal and the original degraded signal are given to the International Telecommunication Union (ITU)-standardized Perceptual Evaluation of Speech Quality (PESQ) estimator to obtain a quality score. Our approach is tested on two different databases consisting of tracheoesophageal speech samples. Results show that our method performs significantly better than the conventional acoustical measures of disordered speech quality.