The choice of the quality features set remains the main issue for the successful speech recognition system. In the literature, quality of features is estimated by calculating the classification error. So that, it is needed to run classification process with each explored feature system in order to choose the highest quality one. Therefore, a major issue of this paper is to propose a methodology for quality establishment of speech features without running the classification process. The proposed methodology is based on metrics that do not need parameters setting, thus the results can be uniformly interpreted across the different problems. The methodology consists of the following parts: 1) establishment of the best metric in combination with used classifier, 2) making a decision regarding the highest quality feature system. In the experiment, we use Dynamic Time Warping (DTW) classifier. The metric of intra/inter class nearest neighbor distances (Q3) is identified as the best one. Employing our proposed methodology, we established Perceptual Linear Prediction analyses to be the highest quality feature system within the explored feature systems. The correctness of the results is confirmed by DTW classification error.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.