A.Ya.Kalyuzhny, V.Yu.Semenov A method of the blind deconvolution of speech signals based on the analysis of line spectral frequencies |
Acoustic bulletin, Vol. 6 ¹ 3, (2003) |
| A problem of the blind deconvolution of speech signals at presence of noise background is considered. An effective method of detection and compensation of media transfer function is proposed. The method is based on the analysis of line spectral frequencies of a distorted signal. Its principal advantage over existing approaches is an adaptive noise compensation inside the structure of the algorithm. An effective procedure of localization of media resonances, based on the analysis of differences of line spectral frequencies, is developed. To diminish the "noise enhancement" effect a block Kalman filter is used. The effectiveness of the resulting method is verified on artificial and real distortions of speech signals. The proposed approach is characterized by lower computational expenses in comparison to the number of modern blind deconvolution of speech signals methods. |
|
KEY WORDS: blind deconvolution, speech signal, media transfer function, block Kalman filtering |
| TEXT LANGUAGE: Russian |