Classification of Persian accents speech using histogram of pitch contour

Classification of Persian accents speech using histogram of pitch contour

Speaker variability significantly impacts the performance of speech recognition. One of the most important facors that cause the variance among speakers is accent. This paper describes an accent recognition system for Persian accents from 5 different dialects. A novel frame work based on histogram of pitch contour of speech is proposed. To reliably detect the accent, a sufficiently long continuous speech segment is required. This research aims to analyze different segmentation approaches of speech signals. SAHAND accent speech data set (SES) is used for data set evaluation. The experimental results confirm that accurate accent recognition is obtained when the speech segments are longer than1.4 second or are composed of more than 15 voiced segments. To recognize the accent in speech, the normalized histogram of pitch frequencies are employed successfully using several approaches, namely K-nearest- neighbor (KNN), Artificial Neural Networks (NNs) based classifier, Naive Bayes and linear discriminant analys(LDA).