Reduced feature sets for vowel recognition

The accuracy of speech recognition systems, to a large extent, depend on the feature sets used for representing the speech data. It has been a continuous process to develop feature sets to perform more accurate speech recognition through ASR (Automatic Speech Recognition) systems. Many feature sets and their different combinations have been tried to achieve better accuracy but the feature set providing completely accurate representation is yet to be formulated. This paper investigates the generation of reduced sets of MFCCs for vowel recognition. The study is more focused on the generation and behavior of MFCCs for different vowel sounds. The goal is to identify the features that enhance their discriminating capabilities and improve the performance of ASR systems for particular sounds. The results of the analysis show that the proposed reduced features perform well and can be further used to improve the accuracy of ASR systems, specifically the ones in resource constrained devices.