Estimating time-varying noise introduced by CVSD for speech enhancement

The authors examine the possibility of improving the quality of speech obtained with continuously variable slope delta-modulation (CVSD), a speech coding method that has been quite widely employed in mobile radios. CVSD is well-known for its robustness and implementation simplicity, but the quality of the resultant speech (commonly at 16 kbps) is often not very satisfactory. Through their study, the authors obtain two crucial findings on the characteristics of the disturbance/noise introduced by CVSD. Based on their findings, they develop a method for estimating the short-time power spectra of CVSD noise, and this enables them to apply existing speech enhancement methods to effectively suppress CVSD noise. Performance assessments based on objective measures such as signal-to-noise ratio (SNR), segmental SNR, and COSH distortion measure, and informal subjective listening tests have all indicated significant improvements in speech quality