Ashraf Alkhairy

Saudi Accented Arabic Voice Bank

The aim of this paper is to present an Arabic speech database that represents Arabic native speakers from all the cities of Saudi Arabia. The database is called the Saudi Accented Arabic Voice Bank (SAAVB). Preparing the prompt sheets, selecting the right speakers and transcribing their speech are some of the challenges that faced the project team. The procedures that meet these challenges are highlighted. SAAVB consists of 1033 speakers speak in Modern Standard Arabic with a Saudi accent. The SAAVB content is analyzed and the results are illustrated. The content was verified internally and externally by IBM Cairo and can be used to train speech engines such as automatic speech recognition and speaker verification systems.

A complex Chebyshev approximation algorithm for FIR filter design

We present an algorithm for computing a real coefficient FIR filter that approximates a desired frequency response. The error criterion used is the complex Chebyshev norm of the weighted error function. The procedure is developed by utilizing the characterization theorem for linear complex Chebyshev approximation. Our algorithm is substantially faster than existing approaches and requires O(N/sup 3/) computations, where N represents the filter length. It has been tested in many applications and has been observed to result in the unique optimal solution.<<ETX>>

On IIR filter design

We present a procedure to solve the IIR filter design problem in the context of magnitude-only frequency response approximation using the tolerance scheme. The method exhibits quadratic convergence to the optimal solution, is robust and numerically stable. It is based on transformation of the specifications to a suitable domain and computation of the optimal solution in it using a modified multiple exchange method.

Design of optimal IIR filters with arbitrary magnitude

We present a method for the computation of an optimal IIR filter that satisfies an arbitrary magnitude frequency response-specified using the tolerance scheme. The method allows for the design of compensators, equalizers, and frequency-selective filters. It exhibits desirable convergence properties and is numerically stable. >

An efficient method for IIR filter design

Presents a method for the design of an IIR digital filter with real coefficients that approximates an arbitrary complex-valued frequency response. The iterative procedure is characterized by desirable convergence properties and has been observed to yield substantially better filters than other approaches. The method has been designed to be robust, easy to implement, widely applicable, and extremely efficient. The computational requirement of the algorithm is O(N/sup 3/) for the design of a filter with N coefficients.<<ETX>>

Synthesis of optimal nonrecursive filters

We present a new method for the synthesis of a minimax-optimal nonrecursive real filter with arbitrary magnitude and phase response. Our approach differs from existing multiple exchange methods, which have been observed to have convergence problems. The method described in this paper is computationally efficient, easy to implement and has been observed to converge to the optimal solution in a large number of cases.<<ETX>>

Estimation of Vocal Tract Resonances Using Spectral Prominent Regions and Artificial Neural Networks

Formants (peaks in the spectrum) are typically found in vowel regions. However, resonances of vocal tract are present in both voiced and unvoiced regions. It is important to focus on extracting vocal tract resonances from speech signal to provide compact and effective representation. This work applies multilayered neural networks for the estimation of vocal tract resonances from linear prediction cepstral coefficients. To train such a neural network, the voiced regions of speech signal are first probed for reliable spectral prominent regions. A sequence of spectral peaks detected by imposing continuity constraints in voiced regions is referred to as reliable spectral prominent regions. A multilayer feedforward neural network is then trained to model a mapping function from linear prediction cepstral coefficients to vocal tract resonances estimated in these reliable spectral prominent regions. This modeled function is used to estimate vocal tract resonances in all speech regions including unvoiced sections. Improvements in vocal tract resonances estimated by this modeled function are 37, 35 and 41%, respectively, for the first three resonances when compared to WaveSurfer, and 49, 32 and 29% for the first three resonances when compared to PRAAT.

Mathematical models of vocal tract with distributed sources

Frequency domain models for the vocal tract with distributed sources are introduced. Pressure and volume velocity distributed sources are modeled within incremental lossy cylindrical pipes in a manner similar to that of modeling electrical components. Equations for transmission matrices of uniform-area vocal tract sections with distributed sources are derived. Transfer functions for pressure and volume velocity at various segments of the vocal tract are computed. Comparison of distributed source models with point source models reveal several improvements on the traditional lumped source modeling method and shows that the effects of the finite impedance constriction and back cavity cannot be adequately modeled using point sources. The distributed source vocal tract framework is important for building articulatory and acoustic models for fricative sounds and other phoneme categories. The paper provides such a mathematical framework for the first time.

Design of recursive digital filters with magnitude specifications

We present an algorithm for the design of an optimal recursive digital filter with a specified magnitude frequency response. The method requires O(N/sup 2/) computations to design a filter of size N, and exhibits numerical stability and quadratic convergence to the optimum within five iterations. The multiple exchange iterative algorithm uses the Chebyshev error criterion in the magnitude-squared frequency response domain, and has been developed using interpolation theory and the alternation theorem for rational function approximation.