Adaptive frequency cepstral coefficients for word mispronunciation detection

Abstract

Systems based on automatic speech recognition (ASR) technology can provide\nimportant functionality in computer assisted language learning applications.\nThis is a young but growing area of research motivated by the large number of\nstudents studying foreign languages. Here we propose a Hidden Markov Model\n(HMM)-based method to detect mispronunciations. Exploiting the specific dialog\nscripting employed in language learning software, HMMs are trained for\ndifferent pronunciations. New adaptive features have been developed and\nobtained through an adaptive warping of the frequency scale prior to computing\nthe cepstral coefficients. The optimization criterion used for the warping\nfunction is to maximize separation of two major groups of pronunciations\n(native and non-native) in terms of classification rate. Experimental results\nshow that the adaptive frequency scale yields a better coefficient\nrepresentation leading to higher classification rates in comparison with\nconventional HMMs using Mel-frequency cepstral coefficients.\n