Dr. Buket D. Barkana, University of Bridgeport
Dr. Inci Saricicek, Eskisehir Osmangazi University, Eskisehir, Turkey
Burak Uzkent, Graduate Student (MS), University of Bridgeport
Automatic noise identification and classification has become a very active subject of research during the last several decades, since it can be directly or indirectly implemented into a very wide area of topics including speech recognition, pattern recognition, and context-aware applications. A large amount of work on environmental noise classification and identification has been completed to separate speech from background noise for robust speech recognition; however a few context-aware applications have attempted to use environmental noise sources. In noise monitoring systems, classification of environmental noises has been provided to help in controlling noise pollution. Adaptive information systems are another research area using environmental noise classification, since the environmental noise can provide a rich source of information about the current context. My research focuses on two things: (i) feature extraction algorithms and (ii) classification methods such as Neural Networks (NN), Support Vector Machines (SVM), K-Means clustering, and Fuzzy Systems.
Our team developed several novel feature sets to classify non-speech environmental sounds. We will continue to do research on this area.
Mel Frequency Cepstrum Coefficients (MFCCs) are considered as a stationary/pseudo-stationary feature extraction method, and they perform very well for the classification of speech and music signals. MFCCs have also been accepted as one of the popular methods to classify non-speech sounds for audio surveillance systems, although MFCCs do not completely reflect the time-varying features of non-stationary non-speech signals. With this research, we introduce a new 2D-feature set which is utilized by a feature extraction method based on the pitch range (PR) of non-speech sounds using the Autocorrelation Function. We compare the classification accuracies of the proposed features to MFCCs by using Support Vector Machines and Radial Basis Function Neural Network classifiers. We have achieved classification accuracies of up to 82.1%, 84.9%, and 89% using PR, MFCC, and PR+MFCC based feature sets, respectively. The proposed feature set provides high accuracy rates as a classifier. Its usage with MFCCs improves the accuracy rates of the given classifiers in the range of 4% to 19.4%, suggesting that they are complementary.
Dr. Buket D. Barkana, University of Bridgeport
Anas Abumunshar, Graduate Student (MS), University of Bridgeport
The goal of this research is to contribute to a richer understanding of accent variation, and its application for speech technology. Emergent of studies on foreign accents has gradually progressed over time. The studies are more prominent today due to large changes in the society. Speaker accent is an important issue in many areas from the robust speaker independent recognition systems to security systems. We started working on this research in the Summer of 2009 as part of Ms. Olagbaju's master thesis. Speech samples (hVd words) of Mandarin and Hindi native speakers of English were recorded in the Summer and Fall of 2009. We are currently analyzing and comparing the acoustical features of the speech signals of non-native and native English speakers.
Dr. Buket D. Barkana, University of Bridgeport
Dr. Saurabh Mukherjee, Banasthali University, India
Dr. Navarun Gupta, University of Bridgeport
Dr. Lawrence V. Hmurcik, University of Bridgeport
This analyzes the expectations and experiences of international graduate students, especially Indian students, in American universities. More specifically, the aim of this study is to find the pre-set goals of international graduate students from India, who are currently pursuing degrees at the graduate level in the School of Engineering at the University of Bridgeport. Volunteers consisting of international graduate students from India pursuing the master in science (MS) in engineering (electrical, computer, mechanical, and technology management) at the University of Bridgeport and undergraduate students in engineering in India were surveyed and are still being surveyed. The research seeks to clarify two broad areas: (1) areas in which student expectations do not match the facilities, resources or practices currently available, and (2) areas where opportunities exist for faculty and staff to meet reasonably held expectations.
Mustafa Cihan Demir (MS), University of Bridgeport
Kerem Ozkan (MS) University of Bridgeport
Muder Almiani (Ph.D. Candidate), University of Bridgeport
Krushika V. Shah, (MS) University of Bridgeport