부산대학교 도서관

In noise suppression methods that are based on an analysis-synthesis approach, speech is enhanced by re-synthesis using acoustic cues extracted from harmonic noise model (HNM) analysis. In this paper, a complete analysis-synthesis framework using HNM is introduced and generalized for different noise conditions. Fine details in choosing and estimating HNM parameters are discussed. Techniques that contribute to robust estimation of HNM parameters are proposed. Evaluation results demonstrate the effectiveness of the proposed HNM-based noise suppression method in low signal-to-noise ratio environments.