Spoken Indian Language Classification Using ANN and Multi-Class SVM
Aarti Bakshi, Sunil Kumar Kopparapu · 2018 International Conference On Advances in Communication and Computing Technology (ICACCT) · 2018
Speech based applications are finding their way into actual use because of the increase in accuracy of ASR due to the increased availability of training corpus and the success of DNN. To build solutions for a multi-lingual country like India the identity of the spoken language is crucial so that an appropriate acoustic and language model can be enabled for the speech recognition to happen seamlessly. In this paper, we looking at spoken Indian language identification we demonstrate the language discriminative power of Artificial Neural Network and Multi-class Support Vector Machine. We conduct a number of experiments on our database of 9 languages. The choices of the languages are a mix of linguistically close and linguistically separated. The performance of the system was analyzed using the well known MFCC features. Results show the superiority in performance of the One vs One multi-class SVM system showing a recognition accuracy of 74% on 9 languages.