A SPEECH REINFORCEMENT METHOD USING SOFT DECISION UNDER FRA-END NOISE ENVIRONMENTS

摘要

PURPOSE: A speech reinforcement method using soft decision under far-end noise environments is provided to remarkably improve the quality of a polluted speech signal by amplifying only clear signals after obtaining a speech signal power through accurate estimation. CONSTITUTION: An LSAP(Local Speech Absence Probability) is obtained, and a clear speech spectrum is estimated by using the LSAP. Through the hear-end background noise estimation, a background noise power is estimated. A speech amplification gain is modified by using the clear speech spectrum and the back noise power. Only clear speech signal is enhanced by using the clear speech spectrum, the background noise power and the voice amplification gain.