Alexander A. Petrovsky

Learn More
This paper addresses the problem of noise estimation for the Karhunen-Loeve transform (KLT) based speech enhancement. The eigenvalues and eigenvectors of the noise covari-ance matrix are tracked using recursive averaging algorithm. This process is controlled by the noise power minima obtained from the noisy signal even during the speech activity periods.(More)
A novel approach to the design and implementation of four-channel paraunitary filter banks is presented. It utilizes hypercomplex number theory, which has not yet been employed in these areas. Namely, quaternion multipliers are presented as alternative pa-raunitary building blocks, which can be regarded as generalizations of Givens (planar) rotations. The(More)
*) This paper presents an alternative factorization for 8-channel general paraunitary filter bank. The utilization of quaternion multiplications leads to a lattice structure being lossless regardless of coefficient quantization. Other advantages are reduced memory requirements and good suitability for FPGA and VLSI implementations. The shown decompositions(More)
*) This paper presents a new, motivated by the theory of hypercomplex numbers, approach to the design of paraunitary filter banks. Quaternion multiplication matrices related to 4D hyperplanar transformations turn out to be usable in the factorization of orthogonal matrices, as an extension and alternative for commonly met Givens rotations. The corresponding(More)
The paper presents a pitch estimation technique based on the robust algorithm for pitch tracking (RAPT) framework. The proposed solution provides estimation of instantaneous pitch values and is not sensitive to rapid frequency modulations. The technique utilizes a different period candidate generating function based on instantaneous harmonic parameters. The(More)
In this paper, we investigate the performance limits of four-channel paraunitary filter banks designed using quaternionic approach. Our aim is to reveal how the maximum achievable coding gain depends on the filter length, linearity of the phase responses, and system one-regularity. We also try to obtain some additional insight into coefficient synthesis for(More)
This paper presents an approach to parametric voice conversion that can be used in real-time entertainment applications. The approach is based on spectral mapping using an artificial neural network (ANN) with rectified linear units (ReLU). To overcome the oversmoothing problem a special network configuration is proposed that utilizes temporal states of the(More)