Skip to Main Content
The variable-order filter method is used in linear prediction (LP) speech modeling. Since the prediction order would be very small for some frames in the method, it can reduce bit rate further. However, due to the large dynamic range of the speech spectrum it might happen that LP analysis is unable to match the higher formants properly if the prediction order is too small. A frequency domain technique is presented to be used in speech modeling to improve the performance of low order linear prediction (LP) in modeling of formants of voiced speech.