GMM Optimization Using Neural Networks for Persian Language Detection

Language identification (LID) in speech signals is an important classification task. In this paper Persian language verification is proposed and evaluated. The system is developed by using Gaussian mixture models as a basic system for tokenizing and a Neural Network as the backend processor. Gaussian Mixture Models can be utilized to model the distribution of feature vector in speech signals for classification. We gathered our language identification corpus from different Satellite TV channels. The results are presented for a system using the GMM Tokenizer in combining with Neural Network. The results of GMM-NN system compared with GMM-Tokenizer system. It is shown that using the Neural Network as the backend processor improves the results significantly.

کلیدواژه ها:

SDC ، en ، LID ، GMM ، Tokenizer ، Language Verification ، Neural network