ABSTRACT
This chapter summarizes fundamentals of speech signal processing and the background of artificial neural network (ANN) application to speech recognition. The period from the 1980s can be divided into two types of contrasting days: the high-tide days of the 1980s, in which ANN-based speech technologies were booming dramatically, and the low-tide days of the latter half of the 1990s, in which people rapidly came to lose research interest in ANN-based approaches. The 1980s were a very productive research period for ANN-based speech processing technologies, even though research was young at that point. The ANN has successfully contributed towards improving the recognition accuracy of preceding conventional systems in many experimental tasks. The scope of such ANN application has been limited to the simple recognition of speech signals that are extracted a priori. One may note that global optimization based on the chain rule of calculus is principally the same as the error back-propagation algorithm, which is an historic ANN-related idea.
