SISTEM IDENTIFIKASI SUARA BERBASIS CONVOLUTIONAL NEURAL NETWORK
Daftar Isi:
- Voice can be utilized as person identification in a biometric system which can be implemented in a security system. Various methods have been proposed to obtain accurate speech identification. The accuracy of such methods were determined by feature extraction. Hence, this study proposes to utilize raw feature. Namely spectrogram which is voice image representation obtained from Short-Time Fourier Transform (STFT). Later, the spectrogram for each sample is processed by Convolutional Neural Network (CNN) as an identifier machine. In this research, the voice is recorded using a microphone. Those data are primary data obtained from 78 students enrolled in the Laboratory of Control and Robotics, Departement of Electrical Engineering, Faculty of Engineering, Sriwijaya University. Each individu records his/her voice 10 times. Thus, there are 780 data in total. Next, 2 types of CNN architecture are utilized, namely simple-CNN architecture and VGG-f architecture. The CNN architecture designed in this study uses the VGG-f model consist of the convolutional layer, pooling layer and softmaxloss as a classification with a parameter size of 224 x 224, a learning rate of 0.001 and a batch size of 256. The results show that the accuracy is 98.7%. It may imply that the combination of spectrogram and CNN may improve the accuracy of speech identification. Keywords : Convolutional Neural Network (CNN), Deep Learning, Voice Recognition, Biometric, Spectogram.