NOISE REDUCTION METHOD USING INSTANTANEOUS SIGNAL-TO-NOISE RATIO FOR OPTIMUM ESTIMATION

摘要

PURPOSE: A noise reduction method using an instantaneous signal-to-noise ratio for optimum estimation is provided to reduce a noise of pattern recognition signals by assuming zero dispersion in terms of an error with respect to a difference among noisy speech, clean speech and the sum of noises. CONSTITUTION: A mapping random variable is defined as a function of a clean signal random variable and a noise random variable. A model parameter describing at least one characteristic of values of the mapping random variable is determined. An estimation value for the clean signal random variable is determined based on the model parameter. The mapping random variable is a signal-to-noise variable.