Artificial neural networks in speech processing: problems and challenges

The author previously (1993, 1996, 1997) designed several new speech processing techniques, including an efficient analysis/synthesis technique named odd sample linear prediction (OSLP), and a speech enhancement (noise cancellation) technique, delayed linear prediction noise cancellation (DLPNC). Traditional digital speech processing approaches have been used while designing these techniques. I conducted a preliminary survey in the field of artificial neural nets (ANN) in speech processing, so that I could apply ANN to these techniques to further improve their performance. While conducting the survey, I realized that the research profile in the area of artificial neural networks is not as mature as in the area of speech processing. Consequently, the research field under consideration (ANN in speech processing) is a relatively young field. This paper summarizes the results of my survey, with special reference to the problems, associated challenges, and limitations I encountered