Wyniki 1-3 spośród 3 dla zapytania: authorDesc:"ALEXANDR PETROVSKY"

Drabinkowa struktura układu mnożenia kwaternionów: analiza i redukcja zakresu dynamicznego

Czytaj za darmo! »

Arytmetyka kwaternionów, czterowymiarowych liczb hiperzespolonych, znalazła ostatnio szereg zastosowań w cyfrowym przetwarzaniu sygnałów [1-10]. W szczególności stanowi ona podstawę nowych transformacji, w których elementarną operacją jest iloczyn kwaternionów, z których jeden jest ustalonym współczynnikiem o wartości bezwzględnej równej jedności. Ponieważ implementacja fundamentalnego działania wpływa na dokładność wyników i wydajność schematów obliczeniowych, które się na nim opierają, celowe jest poszukiwanie nowych ulepszonych algorytmów mnożenia zmiennej hiperzespolonej przez zadany kwaternion jednostkowy. W pracy [11] można znaleźć przegląd znanych algorytmów do obliczania iloczynu kwaternionów, przy czym dużą uwagę zwrócono tam na rozwiązania ukierunkowane na implementa[...]

Using auditory properties in multi-microphone speech enhancement


  An objective of the speech enhancement is to reduce environmental noise while preserving speech intelligibility. In a context of the multi-microphone systems the dereverberation and interference suppression are also expected. Therefore, over the past decades most efforts have been devoted to the beamforming techniques. The key idea of the beamforming is to process the microphone array signals to listen the sounds coming from only one direction. Particularly the noise reduction can be implicitly achieved by avoiding ’noisy’ directions. A linearly constrained minimum variance (LCMV) algorithm has been originally proposed by Frost [1] in the 1970 s and it is probably the most studied beamforming method since then. It minimizes a beamformer output variance subject to the set of linear equations that ensure a constant gain in a specified listening direction. A minimum variance distortionless (MVDR) method [2] can be considered as a special case of the LCVM approach. Similarly, it minimizes a beamformer output variance, but subject to a less restrictive constraint. Another popular technique is a generalized sidelobe canceler (GSC) [3, 4]. It converts the constrained optimization problem defined in the LCVM method into a more efficient, unconstrained form. In addition a processing can be split into two independent stages - the dereverberation and noise suppression, respectively. In order to work reasonably well in the reverberant environments, the classical beamforming techniques often require a system model identification i.e. knowledge of the acoustic room impulse responses or its relative ratios. These parameters can be fixed or estimated adaptively, however in general it is a difficult task. In addition the beamforming methods are usually very sensitive to the system model uncertainties. Recently, much efforts have been made to reformulate the multichannel speech enhancement problem so that the noise reduction can be [...]

Text to speech synthesis system with multi voice capability based on instantaneous voice conversion


  Although text-to-speech (TTS) synthesis is a quite studied issue, researching and adapting new solutions might be still of some importance. Mobile applications deserve special attention since limitations of their performance and capacity constrains from getting superior synthesis quality. Considerable improvements can be achieved, however, through enhancement of the acoustical database rather than the synthesis itself. Despite the fact that many different techniques have been proposed, segment concatenation is still the major approach to speech synthesis. The speech segments (allophones) are assembled into synthetic speech according to phonetic rules of the spoken language. This process involves time-scale and pitch-scale modifications of allophones in order to produce natural- like sounds. The concatenation can be carried out either in time or frequency domain. Most time domain techniques are similar to the Pitch-Synchronous Overlap and Add method (PSOLA) [1]. The speech waveform is separated into shorttime signals by the analysis pitch-marks (that are defined by the source pitch contour) and then processed and joined by the synthesis pitch-marks (that are defined by the target pitch contour). The process requires accurate pitch estimation of the source waveform. Placing analysis pitch-marks is an important stage that significantly affects synthesis quality. Frequency domain techniques deal with frequency representations of the segments instead of their waveforms what requires prior transformation of the acoustic database to frequency domain. The speech segments are often represented as a sum of periodic (deterministic) and noise (stochastic) components as was introduced in [2]. Regarding speech synthesis this model is a very attractive for the following reasons: . explicit control over pitch, tempo and timbre of the speech; . high-quality segment concatenation can be performed using simple linear smoothing laws; . acousti[...]

 Strona 1