期刊名称：International Journal on Smart Sensing and Intelligent Systems
印刷版ISSN：1178-5608
出版年度：2015
卷号：8
期号：1
页码：235-254
出版社：Massey University
摘要：Every individual has some unique speaking style and this variation influences theirspeech characteristics. Speakers’ native dialect is one of the major factors influencing their speechcharacteristics that influence the performance of automatic speech recognition system (ASR). Inthis paper, we describe a method to identify Hindi dialects and examine the contribution of differentacoustic-phonetic features for the purpose. Mel frequency cepstral coefficients (MFCC),Perceptual linear prediction coefficients (PLP) and PLP derived from Mel-scale filter bank (MFPLP)have been extracted as spectral features from the spoken utterances. They are further used tomeasure the capability of Auto-associative neural networks (AANN) for capturing non-linearrelation specific to information from spectral features. Prosodic features are for capturing long -range features. Based on these features efficiency of AANN is measured to model intrinsiccharacteristics of speech features due to dialects.
关键词：Dialect Identification; Auto-associative neural network; Feature compression; Hindi dialects;Spectral and Prosodic features
