SPEECH CODERS BASED ON ANALYSIS-BY-SYNTHESIS TECHNIQUES

摘要

An analysis-by-synthesis speech coder is provided, in which the filter for the spectral weighting of the error signal exploits a more precise human perception model than is conventionally used; thereby resulting in a quality enhanced synthesized speech signal. The weighting filter has a transfer function in which the modulus is represented by the inverse of a global masking threshold of the noise by the speech signal. The thresholds indicate, for the different frequencies in the speech signal, the maximum amount of noise the human ear can tolerate.