Voice recognition system using machine learning techniques Ashraf Tahseen Ali ⇑ , Hasanen S. Abdullah, Mohammad N. Fadhil Department of Computer Science, University of Technology, Baghdad, Iraq article info Article history: Received 30 March 2021 Accepted 5 April 2021 Available online xxxx Keywords: Machine learning Voice recognition Naïve Bayes k-Nearest Neighbor MFCC abstract Voice is a Special metric that, in addition to being natural to users, offers similar, if not higher, levels of security when compared to some traditional biometrics systems. The aim of this paper is to detect impos- tors using various machine learning techniques to see which combination works best for speaker recog- nition and classification. We present several methods of audio preprocessing, such as noise reduction and vocal enhancements, to improve the audios available in real environments. Mel Frequency Cepstral Coefficients (MFCC) are extracted for each audio, along with their differentials and accelerations, to verify machine learning classification methods such as PART, JRip, Nave Bayes, RT, J48, Random Forest, and k- Nearest Neighbor Classifiers. examine the 7 classifiers on two datasets, the extent of accuracy achieved for each classifier. Among the high performance were the random forest algorithm and the naive bias algorithm, and the weak performance of the PART algorithm. Ó 2021 Elsevier Ltd. All rights reserved. Selection and peer-review under responsibility of the scientific committee of the Emerging Trends in Materials Science, Technology and Engineering. 1. Introduction Finger Voice can combine what people say and how they say it by two-factor authentication in a single action. Other forms of identification can help with biometrics, but voice identification is needed for safe and unique authentication. Personal voice recogni- tion and telephone recognition are two variables that can be com- bined with voice [1]. Voice recognition systems are inexpensive and easy to use. In today’s smart world, voice recognition is crucial in a variety of ways. Voice-activated banking, home automation, and voice-activated devices are only a few of the many uses for voice recognition [2]. The process of recognizing a person based on his voice signal is known as speaker recognition. Because of variations in the shape of the vocal tract, the size of the larynx, and other sections of the voice production organs, each person’s sound may be unique [3]. Since voice recognition must be con- ducted in a variety of environments, the features extracted must also be resistant to background noise and sensor mismatches [4]. the speaker’s voice to be used to verify their identity and monitor access to services like voice dialing, telephone banking, dataset access services, information service, voice mail, and security con- trol for sensitive information fields, and remote device access [5]. 2. Literature survey For sixty years, research in automated speech recognition by machines has attracted a lot of interest for a variety of reasons ranging from scientific curiosity about the tools for the mechanical realization of human speech abilities to a request to automate manageable tasks that demand human–machine interactions [6]. In this section, some of the previous work related to this research will be reviewed: In 2017, the researchers have proposed a recognition systems are implemented using both spectro-temporal features and voice-source features. For the i-vector process, classification is per- formed with two separate classifiers, and the accuracy rates are compared. It was decided to compare the efficiency of two separate speaker recognition systems. It is evident from the study that GMM performs better than i-vectors in the case of short utterances, with an accuracy of 94.33%, and that there was a substantial improve- ment in the accuracy rates when concatenated test signals were used [7]. In 2018, the researcher proposed speech recognition sys- tem using SVM. Individual words are separated from continuous speeches using Voice Activity Detection (VAD). Each isolated word’s features were extracted, and the models were successfully educated. Each individual utterance is modelled using SVM. The MFCC is used to describe audio content and is measured as a col- lection of features. By learning from training data, the SVM learn- ing algorithm was used to recognize speech. The proposed audio https://doi.org/10.1016/j.matpr.2021.04.075 2214-7853/Ó 2021 Elsevier Ltd. All rights reserved. Selection and peer-review under responsibility of the scientific committee of the Emerging Trends in Materials Science, Technology and Engineering. ⇑ Corresponding author. E-mail address: ashraf88ashraf888@gmail.com (A. Tahseen Ali). Materials Today: Proceedings xxx (xxxx) xxx Contents lists available at ScienceDirect Materials Today: Proceedings journal homepage: www.elsevier.com/locate/matpr Please cite this article as: A. Tahseen Ali, H.S. Abdullah and M.N. Fadhil, Voice recognition system using machine learning techniques, Materials Today: Proceedings, https://doi.org/10.1016/j.matpr.2021.04.075