Content Details | NUML Online Research Repository

Title EXTRACTING EMOTIONS FROM AUDIO SIGNALS FOR EFFECTIVE REQUIREMENT ENGINEERING

Author(s) Sara Azeem ( MS-CS Scholar ) Supervised by Dr. Sajid Saleem

Abstract Requirements engineering (RE) is a fundamental process in software development project. From the social point of understanding, emotions play a significant role in behavior and also act as developmental motivators. Thus, if RE is considered as a presentation on a set of knowledge-intensive everyday tasks, which include specification, acceptance, rejection and negotiation activities, then the emotional influence, characterizes as a fundamental element in RE. However, the emotional dynamics in RE has not received the consideration it deserves. This research provide collective devotion towards the learning of the emotional content of communication during requirement gathering, and hence, a concrete method has been projected to recognize the emotions within a spoken statement. As speaker emotion recognition is accomplished through processing means that contain isolation of the speech signal, extraction of features, formation of databases and suitable classifiers. Hence this research is an analysis on speaker emotions classification method, addressing central aspects of design of a speaker emotion recognition system. To accomplish this research, a speech emotion recognition (SER) system, which is based on classifier and various techniques for features extraction, is constructed. Mel Frequency Cepstrum Coefficients (MFCC) features are extracted from the audio signals, it was applied in order to catch the most appropriate feature category. Machine learning pattern was used for the emotion classification task. GMM-UBM classifier is used to classify seven emotions. Various databases such as TESS, RAVDEES and SERAD are used for experimental resolution. This research indicates that for TESS database, proposed method achieves an accuracy of 99.63%, for RAVDESS, the accuracy is about 84.37% and for SREAD, the accuracy is 94.05%.

Type Thesis/Dissertation MS

Faculty Engineering and Computer Science

Department Computer Science

Language English

Publication Date 2020-06-29

Subject Machine Learning, Computer Science

Publisher NUML

Contributor(s) Sara Azeem, Dr. Sajid Saleem, Dr. Basit Shehzad

Format PDF

Identifier

Source

Relation

Coverage

Rights NUML, Islamabad

Category MS-CS Thesis Repository

Description