A.Ya.Kalyuzhny, V.Yu.Semenov Computationally efficient method of the speech enhancement based on the block representation of a signal in the state space and a vector quantization |
Acoustic bulletin, Vol. 5 ¹ 3, (2002) |
| A computationally efficient method for enhancement of the speech signals is proposed. Reduction of computations is obtained due to properties of the developed block model of autoregressive signal. A decrease of the filtering error in comparison to the traditional Kalman filter is shown. The problem of estimation of the speech autoregressive parameters at noise background is considered. A two-phase computationally efficient estimation procedure, based on the vector quantization of the autoregressive parameters, is proposed. At the first stage the initial approximation for the optimal set of autoregressive parameters is determined on small number of the autoregressive quantums. Then the value of estimate is improved by efficient iterative procedure. The performance of the resulting method is evaluated on real speech signals. |
|
KEY WORDS: speech signals, signal processing, noise removal, autoregressive model |
| TEXT LANGUAGE: Russian |