Study and analysis of speech emotion recognition

Abstract

This Thesis deals with speech emotion recognition studies in Malayalam Language. newlineMalayalam is one of the South Indian languages and is used by about 32 million people. newlineSpeech emotion recognition is performed in this research work by using spectral, prosodic, newlinewavelet and hybrid features. Three hybrid features are also introduced in this thesis. A newlinemathematical model for expression of emotions is introduced. Speech emotion recognition newlineis explored in this work by using different classifiers. To evaluate the emotion discriminating newlineefficiency of individual features, they are compared with different classifiers and analyzed newlineon the basis of classifier training and testing time. newlineA Malayalam emotional speech database consisting of nine different emotional classes newlineis created for this study. The created database is gender independent. Emotions considered newlinefor collecting the proposed speech corpus were: Neutral, Happy, Sad, Anger, Boredom, newlineFear, Surprise, Calm and Anxiety. A total number of 5 (3 male and 2 female) professional newlineartists participated in the recording phase. All the artists are in the age group of 25 40 newlineyears, and had the professional experience of more than 10 years. The context of database newlineis chosen from a family environment. The collected speech is sampled at 8KHZ frequency newlinerange (4 KHz band limited). All the emotional samples are heard manually to verify their newlineemotional effectiveness and only those words which passed the audibility test are used to newlinestore in the database. A subjective listening test is also conducted by human listeners to newlineevaluate the quality of the database.xvi newlineStudy and Analysis of Speech Emotion Recognition using Spectral, Prosodic and Hybrid Features newlineA preliminary analysis on speech emotion recognition of Malayalam emotional speech newlineis done by using spectral features. This session deals with a brief review on spectral features newlineused for speech emotion recognition