Database for Arabic Speech Commands Recognition

Abstract

Technology is all around us and it’s changing rapidly, expanding Internet access has had huge impacts on everyday lives as people do everything on their phones and computers. The widespread growth in the use of digital computers, have an increasing need to be able to communicate with machines in a simpler manner. One of the main tasks that can simplify communication with machines is speech recognition. In this work, we introduce the Arabic speech commands database that contains six Arabic control order words and Arabic spoken digits. The created database is used to analyze and compare the recognition accuracy and performance of three recognition techniques which are, Wavelet Time Scattering feature extraction with Support Vector Machine (SVM) classifier, Wavelet Time Scattering feature extraction with Long Short-Term Memory (LSTM) classifier, and Mel-Frequency Cepstrum Coefficients (MFCC) feature extraction with K-Nearest Neighbor (KNN) classifier. Finally, the experimental results show that the most accurate prediction of the database commands was 98.1250% given by Wavelet Time Scattering feature extraction and LSTM classifier and the fastest training time for the database was 144 minutes given by MFCC and KNN classifier.