In this correspondence, we propose a compact ellipsoid representation of the reference templates associated with a speaker. The representation captures crucial information about not only the speaker's characteristics, but also the intraspeaker variations. We also derive a from-ellipsoid-surface (FES) distance for use with the ellipsoid representation.

Network structure determination is an important issue in pattern classification based on a probabilistic neural network. In this study, a supervised network structure determination algorithm is proposed. The proposed algorithm consists of two parts and runs in an iterative way. The first part identifies an appropriate smoothing parameter using a genetic algorithm, while the second part determines suitable pattern layer neurons using a forward regression orthogonal algorithm. The proposed algorithm is capable of offering a fairly small network structure with satisfactory classification accuracy.

In this work, we are concerned with optimal estimation of clean speech from its noisy version based on a speech model we propose. We first propose a (single) speech model which satisfactorily describes voiced and unvoiced speech and silence (i.e., pauses between speech utterances), and also allows for exploitation of the long term characteristics of noise. We then reformulate the model equations so as to facilitate subsequent application of the well-established Kalman filter for computing the optimal estimate of the clean speech in the minimum-mean-square-error sense. Since the standard algorithm for Kalman-filtering involves multiplications of very large matrices and thus demands high computational cost, we devise a mathematically equivalent algorithm which is computationally much more efficient, by exploiting the sparsity of the matrices concerned. Next, we present the methods we use for estimating the model parameters and give a complete description of the enhancement process. Performance assessment based on spectrogram plots, objective measures and informal subjective listening tests all indicate that our method gives consistently good results. As far as signal-to-noise ratio is concerned, the improvements over existing methods can be as high as 4 dB.

In this work, we attempt to refine the methods based on autoregressive (AR) modeling for speech enhancement (Paliwal and Basu, 1987); Gibson et al., 1991). As a matter of fact, AR modelling, which is a key strategy of the two above methods, is known to be good for representing unvoiced speech but not quite appropriate for voiced speech which is quite periodic in nature. Here, we incorporate a speech model which satisfactorily describes voiced and unvoiced speeches and silence (i.e., pauses between speech utterances) into the enhancement framework developed in the two above methods, and specifically devise an algorithm for computing the optimal estimate of the clean speech in the minimum-mean-square-error sense. We also present the methods we use for estimating the model parameters and give a description of the complete enhancement procedure. Performance assessment based on spectrogram plots, objective measures and informal subjective listening tests all indicate that our method gives consistently good results.

We investigate whether musical noise, which often exists in speech enhanced using spectral subtraction, can be suppressed. Via exploiting some specific characteristics of human speech, we propose a method that can effectively suppress musical noise without a noticeable effect on speech intelligibility. Performance assessments confirm that our method is effective.

We derive a series of upper and lower bounds for the minimum distance between two noise-free output sequences over unknown channels with fixed-energy finite-duration impulse responses. Using these bounds, we develop a method for determining the minimum distance. Such distance is crucial for determining the worst possible performance of maximum likelihood sequence estimation (MLSE).

We develop a high-resolution method for estimating the directions-of-arrival of partially polarised signals with electromagnetic (EM) vector sensors, of which each provides measurements of the complete electric and magnetic fields induced by EM signals. The method is computationally efficient since unlike many high-resolution methods, it does not involve searching across a multi-dimensional array manifold. In addition, it does not require a priori information about the array system such as sensor positions.

We investigate the linear dependence of the steering vectors of one electromagnetic vector sensor. We show that every 3 steering vectors with distinct DOAs are linearly independent. We also show that 4 steering vectors with distinct DOAs are linearly independent if the ellipticity angles of the signals associated with any 2 of the 4 steering vectors are distinct. We then establish that 5 steering vectors are linearly independent if exactly 2 or 3 of them correspond to circularly polarized signals with the same spin direction. Finally, we demonstrate that given any 5 steering vectors, then for any DOA there exists a steering vector which is linearly dependent on the 5 steering vectors.

Schmidt [1981] has established that ambiguity in steering vectors would lead to ambiguous direction-of-arrival (DOA) estimates. In the literature, arrays free of upto rank-k ambiguities, where k/spl ges/3, have not been identified. The present authors construct a class of cross arrays which are free of upto rank-k ambiguities, where k/spl isin/{1, 2...}. Through the study, they derive a theorem for characterizing inherent ambiguities associated with a class of cross and circular arrays. They also provide a counterexample to the conjecture for characterizing higher rank ambiguity, proposed by Lo and Marple [1992].>