期刊名称：Journal of Theoretical and Applied Information Technology
印刷版ISSN：1992-8645
电子版ISSN：1817-3195
出版年度：2020
卷号：98
期号：4
页码：703-713
出版社：Journal of Theoretical and Applied
摘要：This scientific report illustrates the performance evaluation of the well-known, recently popular neural network Connectionist Temporal Classifier (CTC) for speech recognition. The CTC contains LSTM layers with 256 cells and Momentum Optimizer with learning rate 0.005 and momentum 0.9. Dataset that we have used has 35 native speakers with 360 utterances. For expanding the size of our dataset with overall performance augmentation techniques has been applied using Adobe Audition software, which output 20 more speakers to our original dataset. The result of our experiment has been evaluated with LER (Label error rate). LER measures the inaccuracy between predicted an actual texts. The output of the experiment reported training LER 0.000 and validation LER 0.5.
