It is pointed out that in the analysis of transient signals such as those encounters in speech, or in certain kinds of image processing, standard Fourier analysis is often non satisfactory because the basic functions of the Fourier analysis (sines, cosines, complex exponentials) extend over infinite time, whereas the signals to be analyzed are short-time transients. Reference is made to a method for dealing with transient signals which has recently appeared in the literature. The basis functions are referred to as wavelets, and they utilize time compression (or dilation) rather than a variation of frequency of the modulated sinusoid. Hence, all the wavelets have the same number of cycles. The analyzing wavelets must satisfy a few simple conditions, but are not otherwise specified. There is a wide latitude in the choice of these functions and they can be tailored to specific applications. The wavelets are founded on rigorous mathematical theory, and the expansions are robust. They are applied to detect ventricular delayed potentials (VLP) in the electrocardiogram.>

High-resolution eigenstructure methods developed over the last several years for the passive detection of signals in the far field of a receiving array suffer from the defect of requiring the noise to be white or having a known covariance matrix. This defect can be removed by covariance-matrix subtraction which only requires that the noise be isotropic under rotation of the array. Subtraction results in a received-signal covariance matrix characterized by M-2p zero eigenvalues. There are several likelihood-ratio tests that take advantage of this eigenvalue pattern to determine the number of signal sources. The Cramer-Rao bound for errors in direction estimation indicates that covariance subtraction results in about the same error as other methods using the same amount of received data.

This paper describes our approach to determining the shape of the human vocal tract from the acoustic speech signal. A digital filter model of the vocal tract, which is specified by the locations of six articulatory structures, was examined to determine the set of acoustic signals that can be synthesized. The synthetic acoustic signals were classified by the frequency locations of the first three formant peaks. A sensitivity analysis was performed to determine an appropriate quantization of the vocal tract parameters. A structured table was generated which allowed the vocal tract parameter values to be determined from the set of formant values. A human glide utterance, /ai/, was analyzed to determine the formant locations of contiguous data segments. Using these values, possible vocal tract shape candidates were found from the table. From the total set of candidates occuring over the duration of the utterance, the set of vocal tract shapes that resulted in the minimum parameter variance was determined. It is shown that this minimum variance set, corresponding to a minimum movement of the articulators, produces a reasonable vocal tract shape trajectory for the utterance.

Underwater acoustic communications channels involving surface reflection usually exhibit extensive time and frequency spreading. The phenomenon was studied in a recent paper [5] and the scattering function of the channel was obtained for large Rayleigh parameters using a Fresnel-corrected geometric optics model and a simple onedimensional surface correlation function. It was found that while extensive time spreading and Doppler were indeed predicted by this model, the Doppler shift and time delay were strongly correlated so that the scattering function had the form of a very narrow parabolic ridge. The problem considered in this paper is the extent to which this phenomenon persists when more realistic correlation-function models, dispersion, and small Rayleigh parameters are considered. It is found that the general form of the scattering function is not very much affected by fairly gross changes in the assumed surface statistics, and that substantial correlation between time delay and Doppler can be expected as long as the surface is not completely isotropic. The effect is also observed in a modified form with a small Rayleigh-parameter model, but since this model also features a strong coherent component it may be of less practical significance.

A nonparametric classification procedure based on distribution-free tolerance regions is presented. Without knlowledge of the class probability distributions, the procedure gives information about the expected performance of the classifier through use of only one sample of statistically independent observations from each class. With this procedure, a two-class discriminant can be designed for a given expected false alarm probability or for a given confidence that the false alarm probability is less than a given amount. Three ordering methods are presented that appear intuitively reasonable for minimizing the miss probability. Even though the methods do not, in general, meet this objective, they are easily implemented on a computer and can give good results. A procedure for obtaining a measure of the miss probability is also presented. These methods are applied to the problem of verifying the purported identity of a speaker from a sample of the speaker's voice.

An upper bound on the signal-to-noise ratio (SNR) at the output of an estimator-correlator receiver is derived, and the nature of the transmitted signals and random channels that realize this bound is investigated.

In the above-mentioned paper, Henrichon and Fu present a non- parametric procedure for multivariate, multiclass pattern classification. They obtain a confidence bound on the probability of error through the theory of distribution-free tolerance regions. However, their procedure for obtaining a confidence bound on the basis of a sample from each class is not valid.