查看文章

A dialect identification system for Manipuri using deep-stacked autoencoder-based speech features has been presented in this paper. For this work, Manipuri speech corpora have been used available at Linguistic Data Consortium for Indian Languages (LDC-IL). Mel-frequency cepstrum coefficients (MFCC) were extracted from input speech signals, and then autoencoder (AE) is used for dimensionality reduction. Support vector machines (SVM) and deep neural network (DNN) are used for classification of three major Manipuri dialects, i.e. Sekmai, Kakching and Imphal. Results have shown an average dialect identification performance in terms of accuracy of approximately 70% and 72.77% for SVM and DNN, respectively. The performance of both the classifiers has been compared with and without using autoencoder. The role of stacked autoencoder-based features is found to be significant in Manipuri dialect …