Milan Milosavljević

A statistical pattern recognition approach to robust recursive identification of nonstationary AR model of speech production system

We propose a new robust recursive procedure based on the weighted recursive least squares (WRLS) algorithm with variable forgetting factor (VFF) and frame-based quadratic classifier for identification of nonstationary AR model of speech. Two versions of the frame-based quadratic classifier design procedure are elaborated upon. Experimental results are obtained in analyzing speech signal on voiced and mixed excitation frames.

Estimation of nonstationary AR model using the weighted recursive least square algorithm

A new method of estimating time-varying AR models using weighted recursive least square algorithm with a variable forgetting factor is described. The variable forgetting factor is adapted to a nonstationary signal by a generalized likelihood ratio algorithm through the so-called discrimination function which gives a good measure of nonstationarity. In this way we connect the results from the areas of nonstationary signal estimation and jump detection, and obtain an algorithm which exhibits a good tracking performance together with a high parameter estimation accuracy. The feasibility of the approach is demonstrated with both simulation data and real speech signals.

Adaptive Digital Filters

Adaptive Digital Filters presents an important discipline applied to the domain of speech processing. The book first makes the reader acquainted with the basic terms of filtering and adaptive filtering, before introducing the field of advanced modern algorithms, some of which are contributed by the authors themselves. Working in the field of adaptive signal processing requires the use of complex mathematical tools. The book offers a detailed presentation of the mathematical models that is clear and consistent, an approach that allows everyone with a college level of mathematics knowledge to successfully follow the mathematical derivations and descriptions of algorithms. The algorithms are presented in flow charts, which facilitates their practical implementation. The book presents many experimental results and treats the aspects of practical application of adaptive filtering in real systems, making it a valuable resource for both undergraduate and graduate students, and for all others interested in mastering this important field.

Local echo canceler with optimal input for true full-duplex speech scrambling system

The purpose of this paper is to focus on the local echo-canceling problem of full-duplex scrambled speech communications over a two-wire telephone network when the scrambling transformation is located between the handset and body of a telephone. Such a design makes possible very efficient protection against electromagnetic compromising emanation, which in turn substantially enhances the overall security of a protected communication. We propose a new adaptive FIR filter algorithm for local echo cancellation in such applications. The proposed algorithm differs from the conventional one by the construction of input signals in an optimal way using the D-optimal experiment design. In this way, at each step, we generate a new sample of the D-optimal pilot sequence for the filter parameter estimation. Consequently, the adaptation of the local echo canceler is defined as an initialization process in the first phase of each protected telephone call. The advantage in using the proposed adaptive FIR echo canceler is demonstrated through simulation results.

Time-varying AR speech analysis using robust RLS algorithm with variable forgetting factor

In this paper a new robust recursive method of estimating the linear prediction (LP) parameters of an auto-regressive (AR) speech signal model using weighted least squares (WLS) with variable forgetting factors (VFFs) is described. The proposed robust recursive least squares (RRLS) differs from the conventional recursive least squares (RLS) by the insertion of a suitable chosen nonlinear transformation of the prediction residuals. The RRLS algorithm takes into account the contaminated Gaussian nature of the excitation for voiced speech. In addition, VFF is adapted to a nonstationary speech signal by a generalized likelihood ratio (MGLR) algorithm, which accounts for the nonstationarity of a speech signal. The proposed method has a good adaptability to the nonstationary parts of a speech signal, and gives low bias and low variance at the stationary signal segments.

Quadratic classifier with sliding training data set in robust recursive AR speech analysis

Abstract We propose a robust recursive procedure, based on a weighted recursive least squares (WRLS) algorithm with variable forgetting factor (VFF) and a quadratic classifier with sliding training data set, for identification of non-stationary autoregressive (AR) model of speech production system. Experimental evaluation is done using the results obtained by analyzing speech signal with voiced and mixed excitation frames. Experimental results have shown that the proposed robust recursive procedure achieves more accurate AR speech parameter estimates and provides improved tracking performance.

On evaluating A class of frame-based nonstationary pattern recognition methods using bhattacharyya distance

We consider a possible evaluation of frame-based nonstationary pattern recognition methods by using the upper bound trajectories of the Bayes error based on the Bhattacharrya distance. The experimental part of the work is based on natural speech processing, using isolated spoken Serbian vowels and digits as examples of nonstationary signals. The results obtained justify the use of the upper bound trajectories of the Bayes error expressed by the Bhattacharyya distance as a possible evaluation tool for the class of frame-based nonstationary pattern recognition systems.