In this chapter we present three additional algorithms: ISDCN, BSDCN and FCDCN. The ISDCN and BSDCN algorithms apply a compensation that is a function of the instantaneous SNR exclusively as in the case of SDCN, and they offer a similar recognition accuracy. ISDCN and BSDCN differ from the SDCN algorithm in the way the compensation vectors are estimated, which does not require a stereo database. The FCDCN algorithm is an extension of the SDCN algorithm that offers the highest recognition accuracy of all the algorithms proposed in this monograph, but as in the case of SDCN, it must be recalibrated for every new environment with a stereo database. Finally, we will study the real-time environmental adaptation capabilities of ISDCN, BSDCN and CDCN.
KeywordsRecognition Accuracy Noisy Speech Correction Vector High Recognition Accuracy Warping Path
Unable to display preview. Download preview PDF.