Voice Recognition to Classify “Buka” and “Tutup” Sound to Open and Closes Door Using Mel Frequency Cepstral Coefficients (MFCC) and Convolutional Neural Network (CNN)

Abstract

The consequences of the coronavirus called COVID-19 have been really impactful on society. Many things need to be changed in order to survive this pandemic. People have to avoid physical contact to minimize the probability of getting caught by other people who have been infected. A doorknob has a really big potential to be the medium to spread the virus because the same surface is used by several people. Speech recognition can be used to solve this problem. In this study, Mel Frequency Cepstral Coefficients (MFCC) and Convolutional Neural Network (CNN) are going to be used as the extraction feature and classification method, respectively. We classify the sound signal into two classes (“buka” and “tutup”). People who want to open or close the door just need to say a specific command. This can be helpful to minimize the risk of COVID transmission. A CNN model is developed and fed with an audio file from a curated dataset for training and testing. With this system, we have successfully trained the model with an accuracy of 89% using an epoch of 50 and batch size of 32 as the parameters with a dataset distribution of 8:2 for training and validation. We believe this study will be influential in developing automated door systems using speech recognition, especially in the Indonesian language.