{"title":"The training of feedforward neural network using the unscented Kalman filter for voice classification application","authors":"Zaqiatud Darojah, E. S. Ningrum, D. Purnomo","doi":"10.1109/KCIC.2017.8228451","DOIUrl":null,"url":null,"abstract":"In the previous study, we have investigated that the Extended Kalman Filter (EKF) has the excellennt performance and very fast learning as the training of Feedforward Neural Network (FNN). In the expansion of Kalman filter algorithm for nonlinear estimation, the Unscented Kalman Filter (UKF) was proposed. Enlightened the UKF is superior to EKF, in this study, we investigate the UKF algorithm as the training of FNN for voice classification application. Simulation results show that the UKF has also very excellence performance. The training process need only 2 epochs with the average performance rates in training data is 100% and the average performance rates in the testing data is 94.49%. These results are the same as the EKF-based FNN and the Levenberg-Marquardt Backpropagation but differ in the required training epoch.","PeriodicalId":117148,"journal":{"name":"2017 International Electronics Symposium on Knowledge Creation and Intelligent Computing (IES-KCIC)","volume":"54 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"2017-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"5","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"2017 International Electronics Symposium on Knowledge Creation and Intelligent Computing (IES-KCIC)","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1109/KCIC.2017.8228451","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 5
Abstract
In the previous study, we have investigated that the Extended Kalman Filter (EKF) has the excellennt performance and very fast learning as the training of Feedforward Neural Network (FNN). In the expansion of Kalman filter algorithm for nonlinear estimation, the Unscented Kalman Filter (UKF) was proposed. Enlightened the UKF is superior to EKF, in this study, we investigate the UKF algorithm as the training of FNN for voice classification application. Simulation results show that the UKF has also very excellence performance. The training process need only 2 epochs with the average performance rates in training data is 100% and the average performance rates in the testing data is 94.49%. These results are the same as the EKF-based FNN and the Levenberg-Marquardt Backpropagation but differ in the required training epoch.