I. Yücel Özbek

From acoustics to Vocal Tract time functions

In this paper we present a technique for obtaining Vocal Tract (VT) time functions from the acoustic speech signal. Knowledge-based Acoustic Parameters (APs) are extracted from the speech signal and a pertinent subset is used to obtain the mapping between them and the VT time functions. Eight different vocal tract constriction variables consisting of five constriction degree variables, lip aperture (LA), tongue body (TBCD), tongue tip (TTCD), velum (VEL), and glottis (GLO); and three constriction location variables, lip protrusion (LP), tongue tip (TTCL), tongue body (TBCL) were considered in this study. The TAsk Dynamics Application model (TADA [1]) is used to create a synthetic speech dataset along with its corresponding VT time functions. We explore Support Vector Regression (SVR) followed by Kalman smoothing to achieve mapping between the APs and the VT time functions.

Sleep stage classification based on filter bank optimization

Sleep stage binary classification is studied using single channel EEG signals. The proposed approach is composed of two steps. In the first step, cepstrum coefficients based features are obtained from EEG signals using a filter bank approach which is tuned for sleep stage classification in terms of number of filters and their type. In the second step, these features are used with support vector machine approach for classification. It is observed that obtained results are comparable with the published results, and therefore, it is promising.

Age estimation based on forced expiratory spirometry

This paper examines the benefits of forced expiratory spirometry (FES) test with powerful machine learning algorithms for the purpose age estimation. The proposed method consists of three phases: feature extraction, training of regression models and estimation. Some useful features are determined and extracted from the results of FES test in the first phase. In the second phase, the regression models based on Gaussian Mixture Models (GMM) and Support Vector Machine (SVM) are trained with available training data, and in the final stage, the age of the unknown individuals are estimated by means of trained models. All of the experiments are conducted with a large dataset of 4571 subjects to illustrate the performance gains obtained by the proposed algorithms The average absolute error between the true and the estimated age is 6.54 ± 4.9 (mean ± std.) using GMM method and 6.35 ± 4.7 using DVM method.

Gender prediction based on the expiratory flow volume curve

This study is performed estimated using the gender of the person is the expiration of the current-volume curve obtained from the test. Gender studies estimate is carried out using two different machine learning method. These methods Gaussian Mixture Model (GMM) and Support Vector Machines are (SVM). Gender prediction in both methods are performed using classification. The proposed methods have three main stages. These stages are feature extraction, training and gender of test person is detected. Performance evaluation is made according to the experimental results obtained. As a result of these studies, the gender prediction accuracy of 99.43 per cent are carried out.

Improving the limit of detection (LOD) of microsensor used in detection of brain diseases via wavelet filter

Limit of detection (LOD) gives the concentration amount that a microsensor can detect. It is desirable to have a LOD value of 1μM for microsensors used in brain diseases. The ones that cannot reach this sensitivity value are disposed and cannot be used in the experiments. The goal of this study is to increase the sensitivity of the produced microsensors by decreasing their LOD values. LOD increases linearly by baseline noise. The sensor data is used generally without any baseline filtering in the literature. In this study, LOD values are enhanced 3 times as much by using wavelet filtering, compared with the ones where no filtering is used.

Gender detection with heart sound

In this work, gender detection was carried out using heart sounds of each persons. The proposed method has three stages to make detection gender with heart sound. First, some features are obtained from heart sounds. Second, modelling is made by Gauss Mixture Model (GMM) by using obtained these features and models are trained. Finally, tested person is decided to be he or she with likelihood ratio test.

Microelectrod fabrication for diagnosis and treatment of brain disorders

In this study, it is aimed to produce microelectrodes which can be used in the detection of neurotransmitters that are related with brain disorders such as Parkinson, Epilepsy, and Schizophrenia and that exist in the central nervous system (CNS). A 4-channel, ceramic-based fabrication is performed towards this goal by using photolithographic methods. The time-current graphic response against the addition of H2O2 the produced microelectrode is analyzed in the calibration test. It is observed that the response is in stepwise form. In addition, limit of detection (LOD) of the produced microelectrodes and linearity values are shown to be within the desired ranges.

Locus equations for Turkish plosives

In this study, the locus equations of Turkish plosives are examined. The speech database including all the combinations of Turkish plosives and vowels is prepared for this purpose. The formant frequencies of plosives and successive vowels are manually determined for the speech data in the database. The locus equations for Turkish plosives are determined by using this data. The observations show that the parameters of locus equations can be a very distinctive feature for Turkish plosives.

Audiovisual articulatory inversion based on Gaussian Mixture Model (GMM)

In this study, we examined articulatory inversion using audiovisual information based on Gaussian Mixture Model (GMM). In this method the joint distribution of the articulatory movement and audio (and/or visual) data are modelled via a mixture of Gaussians. The conditional expected value of the GMM is used as regression function between the audio (and/orvisual) and ar-ticulatory spaces. We also examined various fusion methods in order to combine acoustic and visual information in articula-tory inversion. The fusion methods improve the performance of articulatory inversion.