Skip to Main Content
This paper presents speech and a noise separation methods for achieving high quality of speech coding. In speech separation, a pitch harmonics restoration method is proposed. This method can effectively suppress the so-called musical noise and reduce speech distortion. In noise separation, the noise-base estimated in the speech separation process is used as a separated background noise and a method for encoding the noise with low bit rates is proposed. The proposed methods used as a preprocessor of the adaptive multi-rate wideband (AMR-WB) coder are evaluated by the degradation category rating (DCR) test. An average of 0.3-point improvement in performance under the noise conditions is achieved compared with the conventional method without using speech and noise separations.