Kendali Pintu Air Otomatis Berbasis Speech Recognition Menggunakan Metode MFCC dan Jaringan Syaraf Tiruan

Abstrak

In this research, the speech of command for sluice recorded and then the coefficient of signal speech taken using MFCC algorithm. These coefficients becomes features of speech signal and using as input in neural network to be training data. After training, the testing using data test conducted to see how the algorithm of neural network have good working and how the effectiveness of controlling automatic sluice. There is 5 coefficients of the word of “BUKA” (English: “OPEN”) and so is word of “TUTUP” (English: “CLOSE”). That 5 coefficients was taken as features then repeated 15 times to be used as training data in neural network. The total number of command speech is 30. The 20 test data for “BUKA” and 20 for “TUTUP” used to test the system in controlling sluice by speech. After the test we get the result the accuracy for “BUKA” is 75% and 55% for “TUTUP”. Testing with giving command using “OPEN” and “CLOSE” give us the result by there is no stability in output of the value of neural network (y). From the neural network training, the output of training is 1.01325 for “BUKA” and 0.0930902 for “TUTUP”. With using the other word we get the value of y is changeable inconstantly.