Adaptive Spectral Shaping Using LP Filter Decomposition to Improve Perceptual Quality of Coded Speech

摘要

An adaptive time-domain postfiltering technique based on the synthesis LP filter factorisation is proposed. Information is gathered about the relation between the LP filter poles and formants for this factorisation. This technique shapes the main formant differently from the other formants. Pole locations representing the main formant are modified and optimum shaping constants for each formant are searched to make more narrower main formant bandwidth while maintaining other formant information and more attenuation in valley regions.