Our purpose is to design a useful tool which can be used in psy- chology to automatically classify utterances into five emotional states such as anger, happiness, neutral, sadness, and surprise. The major contribution of the paper is to rate the discriminating capa- bility of a set of features for emotional speech recognition. A total of 87 features has been calculated over 500 utterances from the Danish Emotional Speech database. The Sequential Forward Se- lection method (SFS) has been used in order to discover a set of 5 to 10 features which are able to classify the utterances in the best way. The criterion used in SFS is the crossvalidated correct clas- sification score of one of the following classifiers: nearest mean and Bayes classifier where class pdfs are approximated via Parzen windows or modelled as Gaussians. After selecting the 5 best fea- tures, we reduce the dimensionality to two by applying principal component analysis. The result is a 51.6% ± 3% correct classifi- cation rate at 95% confidence interval for the five aforementioned emotions, whereas a random classification would give a correct classification rate of 20%. Furthermore, we find out those two- class emotion recognition problems whose error rates contribute heavily to the average error and we indicate that a possible reduc- tion of the error rates reported in this paper would be achieved by employing two-class classifiers and combining them.

In this paper, the expectation-maximization (EM) algorithm for Gaussian mixture modeling is improved via three statistical tests. The first test is a multivariate normality criterion based on the Mahalanobis distance of a sample measurement vector from a certain Gaussian component center. The first test is used in order to derive a decision whether to split a component into another two or not. The second test is a central tendency criterion based on the observation that multivariate kurtosis becomes large if the component to be split is a mixture of two or more underlying Gaussian sources with common centers. If the common center hypothesis is true, the component is split into two new components and their centers are initialized by the center of the (old) component candidate for splitting. Otherwise, the splitting is accomplished by a discriminant derived by the third test. This test is based on marginal cumulative distribution functions. Experimental results are presented against seven other EM variants both on artificially generated data-sets and real ones. The experimental results demonstrate that the proposed EM variant has an increased capability to find the underlying model, while maintaining a low execution time.

Two approaches for ultrasonic image processing are examined. First, signal-adaptive maximum likelihood (SAML) filters are proposed for ultrasonic speckle removal. It is shown that in the case of displayed ultrasound (US) image data the maximum likelihood (ML) estimator of the original (noiseless) signal closely resembles the L(2) mean which has been proven earlier to be the ML estimator of the original signal in US B-mode data. Thus, the design of signal-adaptive L(2) mean filters is treated for US B-mode data and displayed US image data as well. Secondly, the segmentation of ultrasonic images using self-organizing neural networks (NN) is investigated. A modification of the learning vector quantizer (L(2 ) LVQ) is proposed in such a way that the weight vectors of the output neurons correspond to the L(2) mean instead of the sample arithmetic mean of the input observations. The convergence in the mean and in the mean square of the proposed L(2) LVQ NN are studied. L(2) LVQ is combined with signal-adaptive filtering in order to allow preservation of image edges and details as well as maximum speckle reduction in homogeneous regions.