Abstract

Speech enhancement has been used in a variety of applications such as mobile phones, hearing aids and speech recognition systems. One of the most fundamental steps in manyspeech enhancement methods is noise estimation. Inexact noise estimation results in undesired effects in the enhanced signal. While over-estimating the noise leads to speech distortion, noise under-estimation leaves some annoying noise in the enhanced signal. Recently, minimum-statistics and soft-decision noise estimation methods have been paid high attention. In this paper, we evaluate these methods and compare their performance in speech+noise conditions.

Key words: Noise estimation, speech enhancement, minimum statistics, soft decision.