Robustní parametrizace řeči ve frekvenční oblasti

Robust Speech Parameterization in Frequency Domain

Project is bent on speech signal processing, its parametrization and using this parameters in robust speech recognition by HMM net. Main part to use non-linear spectral smoothing, gain detection of this smoothing on recognizer robustness, compare this method with another used method and possibility to use this algorithm in recognizers with more inner states nets.

Speech signal, parametrization, HTK toolkit, signal processing, cepstrum, HMM recognition, robust recognizer, MFCC, speech recognition, non-linear spectral smoothing, RASTA filter

ŠEBESTA, Jaroslav. Robustní parametrizace řeči ve frekvenční oblasti. Brno, 2006. Year Project. Brno University of Technology, Faculty of Information Technology. 2006-05-10. Supervised by Motlíček Petr. Available from: https://www.fit.vut.cz/study/thesis/4572/