期刊名称：International Journal of Electronics, Communication and Soft Computing Science and Engineering
印刷版ISSN：2277-9477
出版年度：2014
卷号：3
期号：Special
出版社：IJECSCSE
摘要：Speech is the fundamental, most effective, reliable andcommon medium to communicate in real time systems. There areso many applications of speech still to be far from reality justbecause of lack of efficient and reliable noise removal mechanismand techniques for preserving or improving the intelligibility forthe speech signals. In this paper attempt has been steppedtowards surveying the methodologies for soft computing basedspeech recognition techniques for speech enhancement inmultimedia applications. Because of their many applications andtheir relative ease of implementation, single-channel speechenhancement algorithms have received much attention. As aconsequence, a vast amount of publications on estimationprocedures and their implementation in noise reduction systemsexists. However, there has been little systematic research on thetheoretic performance of such estimators. Missing data techniques(MDTs) have been widely employed and shown to improve speechrecognition results under noisy conditions. This paper presentsnew technique which improves upon previously proposed sparseimputation techniques relying on the least absolute shrinkage andselection operator (LASSO). LASSO is widely employed incompressive sensing problems. However, the problem withLASSO is that it does not satisfy oracle properties in the event ofa highly collinear dictionary, which happens with featuresextracted from most speech corpora
