Christos Boukis

Collaborative Adaptive Learning using Hybrid Filters

A novel stable and robust algorithm for training of finite impulse response adaptive filters is proposed. This is achieved based on a convex combination of the least mean square (LMS) and a recently proposed generalised normalised gradient descent (GNGD) algorithm. In this way, the desirable fast convergence and stability of GNGD is combined with the robustness and small steady state misadjustment of LMS. Simulations on linear and nonlinear signals in the prediction setting support the analysis.

A novel algorithm for the adaptation of the pole of Laguerre filters

This letter proposes a novel stochastic gradient algorithm for the online adaptation of the pole position in Laguerre filters. The proposed algorithm exploits the inherent relationship between the values of the filter coefficients and the value of the Laguerre pole. This leads to an unbiased solution and, hence, a more accurate estimate of the error gradient. Simulations in a system identification setting support the analysis

Toward bias minimization in acoustic feedback cancellation systems

A novel technique for bias suppression within acoustic feedback cancellation systems is proposed. This is achieved based on the use of all-pass filters in the forward part of the hearing aid. The poles of these filters are made time-varying, which results in a frequency response with constant magnitude and varying phase. This is a desired feature of the proposed approach, since the results from human psychoacoustics show that the human ear is not sensitive to moderate phase perturbations. The derivation of the proposed algorithms for the time variation of the location of the poles of all pass filters is based on a rigorous analysis of the phenomenon of bias in acoustic systems. Practical issues, such as the dependence of the steady-state error on the order of the all-pass filter, the number of varying poles, and their standard deviation are examined and strategies for the variation of the poles are introduced. Results obtained from a simulated hearing aid are provided to support the analysis. The quality of the processed audio signals is evaluated through subjective tests.

Artificial Intelligence and Innovations 2007: from Theory to Applications

Most recent volume from the AIAI conference series International Federation for Information Processing The IFIP series publishes state-of-the-art results in the sciences and technologies of information and communication. The scope of the series includes: foundations of computer science; software theory and practice; education; computer applications in technology; communication systems; systems modeling and optimization; information systems; computers and society; computer systems technology; security and protection in information processing systems; artificial intelligence; and human-computer interaction. Proceedings and post-proceedings of referred international conferences in computer science and interdisciplinary fields are featured. These results often precede journal publication and represent the most current research. The principal aim of the IFIP series is to encourage education and the dissemination and exchange of information about all aspects of computing.

Convex Combination of Multiple Statistical Models With Application to VAD

This paper proposes a robust voice activity detector (VAD) based on the observation that the distribution of speech captured with far-field microphones is highly varying, depending on the noise and reverberation conditions. The proposed VAD employs a convex combination scheme comprising three statistical distributions - a Gaussian, a Laplacian, and a two-sided Gamma - to effectively model captured speech. This scheme shows increased ability to adapt to dynamic acoustic environments. The contribution of each distribution to this convex combination is automatically adjusted based on the statistical characteristics of the instantaneous audio input. To further improve the performance of the system, an adaptive threshold is introduced, while a decision-smoothing scheme caters to the intra-frame correlation of speech signals. Extensive experiments under realistic scenarios support the proposed approach of combining several models for increased adaptation and performance.

Voice activity detection using audio-visual information

An audio-visual voice activity detector that uses sensors positioned distantly from the speaker is presented. Its constituting unimodal detectors are based on the modeling of the temporal variation of audio and visual features using Hidden Markov Models; their outcomes are fused using a post-decision scheme. The Mel-Frequency Cepstral Coefficients and the vertical mouth opening are the chosen audio and visual features respectively, both augmented with their first-order derivatives. The proposed system is assessed using far-field recordings from four different speakers and under various levels of additive white Gaussian noise, to obtain a performance superior than that which each unimodal component alone can achieve.

A normalized mixed-norm adaptive filtering algorithm robust under impulsive noise interference

A normalized robust mixed-norm (NRMN) algorithm for system identification in the presence of impulsive noise is introduced. The standard robust mixed-norm (RMN) algorithm, despite its ability to cope with impulsive noise by virtue of combining the first and second error norm in the cost function it minimizes, exhibits slow convergence, requires a stationary operating environment, and employs a constant step-size which needs to be determined a-priori. To overcome these limitations, the proposed NRMN algorithm introduces a time varying learning rate which is derived based upon the dynamics of the input signal, and thus no longer requires a stationary environment, a major drawback of the RMN algorithm. The normalized step-size is bounded from above and a parameter is introduced within its upper-bound, which provides a trade-off between the convergence rate and the steady-state coefficient error. The analysis and experimental results show that the proposed NRMN exhibits increased convergence rate and substantially reduces the steady-state coefficient error, as compared to the least absolute deviation (LAD) and RMN algorithms.

Automated sleep breath disorders detection utilizing patient sound analysis

Abstract Results of clinical studies suggest that there is a relationship between breathing-related sleep disorders and behavioral disorder and health effects. Apnea is considered one of the major sleep disorders with great accession in population and significant impact on patients health. Symptoms include disruption of oxygenation, snoring, choking sensations, apneic episodes, poor concentration, memory loss, and daytime somnolence. Diagnosis of apnea and breath disorders involves monitoring patients biosignals and breath during sleep in specialized clinics requiring expensive equipment and technical personnel. This paper discusses the design and technical details of an integrated low-cost system capable for preliminary detection of sleep breath disorders at patients home utilizing patient sound signals. The paper describes the proposed architecture and the corresponding HW and SW modules, along with a preliminary evaluation.

A class of stochastic gradient algorithms with exponentiated error cost functions

A novel class of stochastic gradient descent algorithms is introduced based on the minimisation of convex cost functions with exponential dependence on the adaptation error, instead of the conventional linear combinations of even moments. The derivation is supported by rigourous analysis of the necessary conditions for convergence, the steady state mean square error is calculated and the optimal solutions in the least exponential sense are derived. The normalisation of the associated step size is also considered in order to fully exploit the dynamics of the input signal. Simulation results support the analysis.

A modified Armijo rule for the online selection of learning rate of the LMS algorithm

The use of the Armijo rule for the automatic selection of the step size within the class of stochastic gradient descent algorithms is investigated, and the Armijo rule learning rate least mean square (ALR-LMS) algorithm is introduced. This algorithm is derived by integrating an appropriately modified version of the Armijo line search to the least mean square filter update. The analysis of the stability, robustness and the bounds on the parameters which guarantee convergence is conducted, and some practical issues relating the choice of parameters of the ALR-LMS and computational complexity are addressed. Comprehensive simulation results in the system identification and prediction setting support the analysis.