ANALYZING ACOUSTIC MARKERS OF EMOTION IN ARABIC SPEECH

Abstract

Abstract ANALYZING ACOUSTIC MARKERS OF EMOTION IN ARABIC SPEECH Muna Bin Othman A dissertation submitted to the University of Manchester for the degree of Master of Philosophy (MPHIL), 2017 This study aims to obtain detailed acoustic knowledge of how speech is modulated when a speaker’s emotion changes from neutral to certain emotional states based on measurements of acoustic parameters related to speech prosody. This can be used effectively in many applications of synthesis or recognition systems of emotions in Arabic speech. The common problems often faced by studying emotions in Arabic speech are explored, including the complexity of the phonetic rules and diacritic system of Arabic which makes Arabic speech harder to process than for most other languages, and a lack of freely available emotional corpora for Arabic speech. The acoustic features of pitch, intensity and duration are extracted from a small corpus and then used to classify the four emotions: neutral, happy, sad and anger in Arabic speech. A range of experiments are conducted in order to identify and investigate the key features of emotional speech in Arabic and to determine which one has the most effect on the emotional speech and to use them to classify the emotions. The results of the experiments are analyzed. The main findings are that using the combination of these features enhanced the performance and Anger was the most challenging to identify while other emotions were classified in different range of accuracy. Finally, several suggestions for future experiments in this area are discussed.

Date of Award	1 Aug 2019
Original language	English
Awarding Institution	The University of Manchester
Supervisor	Allan Ramsay (Supervisor)