Malaysian English accents identification using LPC and formant analysis

摘要: In Malaysia, most people speak several varieties of English known as Malaysian (MalE) and there is no uniform version because the existence multi-ethnic population. It a common scenario that Malaysians particular local Malay, Chinese or Indian accent. As commercial speech recognizers have been developed using standard language, it challenging task for achieving highly efficient performance when other accented are presented to this system. Accent identification (AccID) can be one subsystem in speaker independent automatic recognition (SI-ASR) system so deterioration issue its tackled. paper, important features three ethnic groups MalE speakers extracted Linear Predictive Coding (LPC), formant log energy feature vectors. subsequent stage, accent identity predicted K-Nearest Neighbors (KNN) classifier based on information. Prior, preprocessing parameters LPC order investigated properly extract features. This study conducted small set corpus pilot determine feasibility AccID which has never reported before. The experimental results indicate promising accuracy 94.2% upon fusion sets LPC, formants energy.