Speaker State Classification Using Machine Learning Techniques

Abstract

Speech Emotion Recognition, Speech Recognition, Speech Processing are the synonyms that deal with the speech signals vibrated from the spoken words. Speech is natural way to interact with humans but it doesn’t necessarily apply while interacting with a computer. Now-a-days users interact with smart phones through their verbal language and this is possible because of Natural Language Processing and Speech Processing. Through Speech Processing we can extract information through speaker state classification such as drunken, drowsy, gender, age, emotions, language identification etc. The main content of paper deals with the speaker state classification whether a person is alcoholic (A), non-alcoholic (NA) or completely non-alcoholic (CNA). The database used is Alcohol Language Corpus (ALC). From the database two types of features are extracted midterm features and short term features. Considering midterm features classifiers such as Support Vector Machine, k Nearest Neighbor, Random forest, Gradient Boosting and Extra Trees are applied. The results are then visualized to classify the most alcoholic from non-alcoholic. The comparative study of all these classifiers can be done effectively.