Nonlinear predictive analysis of speech by iterative approach

The filter involving the adaptation scheme of Volterra Series Least Mean Square(VSLMS) algorithm is a representative adaptive nonlinear filter, which has been applied to a lot of engineering applications. However, when the VSLMS filter is used as an adaptive predictor of speech, a large number of speech data samples are required to minimize the predictive error. And if the VSLMS predictor is used for short-term prediction with a high order of the quadratic kernel to increase the predictive gain, it is suffered from its numerical unstability. To conquer such problems, an iterative approach is proposed in this paper. The iterative approach gives an effect to utilize a large number of speech data samples by using a segmented speech signal repeatedly. Experiments are conducted on continuous speech and it is shown that the predictive accuracy of the VSLMS predictor is improved by relying on the iterative approach.