Clustering based emotional speech recognition using fuzzy and K-means for tamil language

• Main Authors: John, Bennilo Fernandes, Kusumanchi, T. P. S. Kumar, Ramamoorthi, Agilesh Saravanan, Ramanathula, Sireesha, Kongala, Raju
• Format: Conference Proceeding
• Language: English
• Subjects: Algorithms; Clustering; Emotion recognition; Emotions; Speech recognition
• ISSN: 0094-243X; 1551-7616
• DOI: 10.1063/5.0213359

Communication among people and robots is still a difficult task where the machine should understand and respond to the manner in which it interacts like emotions, so that interaction between humans and computers seems simpler and natural. In this article, we examine a method for classifying the Tamil emotional database through a comparative analysis of three Fuzzy prototype techniques: FCM, KFCM, and k-means algorithm. The goal is to identify the most efficient approach. The extraction of the emotional speech feature is analyzed via MFCC delta & Spectral Skewness concatenation methodology to get more features for the analysis. For training and testing purposes, both male and female speech samples an emotional Tamil language database was developed with PCA & followed by ICA technique in order to utilize the data for higher order dataset. Emotion detection outcomes have both the benefits and drawbacks of their own approach. Our analysis describes the efficiency of the KFCM methodology shows more than k-means and FCM approaches and analysis it’s clear that the emotions like anger, happy and normal shows higher rate of accuracy than other emotions with overall approximation of 84% of precision and also the time taken for the execution by KFCM is 0.238 mins, where the other methods take slightly higher time for execution.