Improving Voice Assistant System Performance Using Machine Learning Technique

Abstract

The purpose of this research is to improve the voice assistant systems capability from user dictation to recommendations receiving stage by normalizing observed output latencies across various levels of utterances. To achieve this, voice assistant client data will be subjected to online or offline voice recognition types on basis of reference to classification output achieved by machine learning models trained on datasets created from voice assistant system usability aspects. Statistical analysis has been done on the dataset to determine applicable machine learning models selection. Due to the multi-class nature of the dataset, multiclass logistic regression, KNN model, and Naive Bayes were chosen for building a classification model and comparing efficiencies. Naive Bayes resulted in better accuracy while compared to Logistic but similar to KNN Model. An improved system design approach is presented at the end of the study.