A Bayesian classifier may supply an optimal estimate of the a posteriori class probabilities for classifying stochastic patterns, provided that the underlying statistical model of the problem is known. In the absence of such a priori knowledge, one valuable alternative is the Boltzmann perceptron classifier (BPC), a statistical neural based classifier, which was shown to have the capability of Bayesian like decisions. The original learning algorithm of the BPC requires a knowledge of the a posteriori probabilities for the given training set. However, these probabilities are seldom known in advance, and instead, labeled training data is given for which only the class membership associated with each training sample is known. The authors introduce a regulated learning scheme which estimates the class probabilities from such labeled data and constructs a classifier that generalizes well for new data.>

A technique for efficient vector quantization (VQ) design based on conjugate-gradient search methods is introduced. It is demonstrated experimentally that the new design algorithm performs better than the generalized Lloyd algorithm in many cases in terms of computation time and/or quality of the codebook it produces.>

It is shown that neural network architectures may offer a valuable alternative to the Bayesian classifier. With neural networks, the a posteriori probabilities are computed with no a priori assumptions about the probability distribution functions (PDFs) that generate the data. Rather than assuming certain types of PDFs for the input data, the neural classifier uses a general type of input-output mapping which is then designed to optimally comply with a given set of examples called the training set. It is demonstrated that the a posteriori class probabilities can be efficiently computed by a deterministic feedforward network which is called the Boltzmann perceptron classifier (BPC). Maximum a posteriori (MAP) classifiers are also constructed as a special case of the BPC. Structural relationships between the BPC and a conventional multilayer perceptron (MLP) are given, and it is demonstrated that rather intricate boundaries between classes can be formed even with a relatively modest number of network units. Simulation results show that the BPC is comparable in performance to a Bayesian classifier.>

An accurate, robust, and reliable pitch detection scheme is outlined. The algorithm extracts the pitch with a very high resolution (i.e., as a real number) despite the finite (8-kHz) resolution of the sampled speech sequence. This makes it possible to analyze the pitch variation process in disarthric patients for medical purposes and to develop an accurate synchronous spectral analysis scheme. The computational complexity of the proposed algorithm is well below the capacity of modern digital signal processing technology and therefore can be implemented in real-time.>

A point process model of the speech excitation wave for sustained vowels has been formulated, where the pitch is modulated by a physiological disturbance (vocal tremor). It has been demonstrated that the power spectrum of an impulse train, representing the glottal pulses, is composed of a periodic replication of the spectrum of the modulating signal, accompanied by an impulse train at the carrier frequency. Hence, the power spectrum of the vocal tremor can be estimated from the pitch. A trend location algorithm has been developed to locate and remove local trends from the signal prior to the evaluation of the pitch power spectrum. Thus, spectral representation of the speech excitation enables estimation of vocal tremor.>

An efficient short-time spectral analysis scheme of periodic, or quasi-periodic signals, such as voiced speech, is outlined. Generally, the signal is sampled at a fixed sampling rate irrespective of the signal period. As a consequence, the resulting sampled sequence may not preserve the periodic structure. The proposed scheme is based on an implicit reconstruction of the signal from its aperiodic sampled sequence followed by a pitch-synchronous resampling of the reconstructed signal. Applying the FFT algorithm to the interpolated sequence for single periods is shown to give the desired Fourier coefficients corrupted by an aperiodic high frequency noise. However, the fact that this noise is correlated to the signal is used to obtain a closed-form solution for retrieving the desired Fourier coefficients from the noisy values. The proposed scheme is also demonstrated for simulated data.