Javanese Gender Speech Recognition Using Deep Learning And Singular Value Decomposition

Speech detection is an interesting research field. Research in speech recognition uses a variety of models that aim to improve accuracy, one of which is by using Deep Learning, but high dimensional data problems are one of the problems that cause a decrease in the quality of speech recognition accuracy. This paper discusses the gender voice recognition of Javanese people who are processed using the Mel Frequency Cepstral Coefficient (MFCC) extraction feature, then voice classification is done using the Deep Learning method combined with the Singular Value Decomposition (SVD) method in reducing the dimensions of the data produced. By using a 70% split ratio for training data and 30% for testing data the results of the research show that the Deep Learning method's accuracy is 97.78% higher than the Logistic Regression method of 95.56% and SVM of 93.33%. Speech recognition research shows that the Deep learning and SVD method can be used in performing speech recognition with a high degree of accuracy.