Nonlinear Prediction of Speech Signal using Short Term Forecasting Algorithm

This research paper presents nonlinear prediction of speech signal using forecasting algorithm. Here both clean and noisy speech samples are predicted using the nearest neighbors of the sample to be predicted. We also discuss the optimum number of nearest neighbors required to obtain the best results. Various parameters like embedding dimension, delay and number of nearest neighbors were varied to analyze the optimal parameter set for a chaotic predictor for speech signals. It was observed that dimensionality of speech varies between 6 and 10 and the prediction error was reducing with the increase of number of nearest neighbors.