Time series prediction method for stochastic acoustic signals by the use of an adaptive function model

Abstract In principle, some kinds of correlations can be found between the past data and the future value in the actually measured acoustic signals. This paper describes a new trial of predicting the fluctuation of a stochastic acoustic signal by extracting the information on many correlation properties from its measured past data. More explicitly, the prediction algorithm is proposed in a general form of series expansion type with a linear combination of newly introduced adaptive functions, with the use of a generalised error evaluation criterion. Finally, the validity and effectiveness of the proposed prediction method have been confirmed by a computer simulation and an application to the actual stochastic acoustic signal measured near a national main road in Hiroshima City.