Digital processing of speech signals. / Lawrence R. Rabiner, Ronald W. Schafer

The formal organization of this book is as follows. Chapter 1 provides an introduction to the área of speech processing, and gives a brief discussion of applica-tion áreas which are directly related to topics discussed throughout the book. Chapter 2 provides a brief review of the fundamentáis of digital signal processing. It is expected that the reader has a solid understanding of linear systems and Fourier transforms and has taken, at least, an introductory course in digital signal processing. Chapter 2 is not meant to provide such background, but rather to establish a notation for discussing digital speech processing, and to provide the reader with handy access to the Rey equations of digital signal processing. In addition, this chapter provides an extensive discussion of sampling, and decimation and interpolation, key processes that are fundamental to most speech processing systems. Chapter 3 deals with digital modeis for the speech signal. This chapter discusses the physical basis for sound production in the vocal tract, and this leads to various types of. digital modeis to approximate this process. In addition this chapter gives a brief introduction to acoustic phonetics; that is, a discussion of the sounds of speech and some of their physical properties. Chapter 4 deals with time domain methods in speech processing. Included in this chapter are discussions of some fundamental ideas of digital speech processing- e.g., short-time energy, average magnitude, short-time average zero-crossing rate, and short-time autocorrelation. The chapter concludes with a section on a nonlinear smoothing technique which is especially appropriate for smoothing the time-domain measurements discussed in this chapter. Chapter 5 deals with the topic of direct digital representations of the speech waveform-i.e., wavefonn coders. In this chapter the ideas of instantaneous quantization (both uniform and nonuniform), adaptive quantization, differential quantization, and predictive coding (both fixed and adaptive) are discussed and are shown to form the basis of a variety of coders from simple pulse code modulation (PCM) to adaptive differential PCM (ADPCM) coding, I Chapter 6 is the first of two chapters that deal with spectral representations of | speech. This chapter concerns the ideas behind short-time Fourier analysis and synthesis of speech. This área has traditionally been the one which has received most attention by speech researchers since some of the key speech processing systems, such as the sound spectograph and the channel vocoder, are directly related to the concepts discussed in this chapter. Here it is shown how a fairiy general approach to speech spectral analysis and synthesis provides a framework for discussing a wide variety of speech processing systems, including those mentioned above. Chapter 7, the second chapter on spectral representations of speech, deals with the área of homomorphic speech processing. The idea behind homomorphic processing of speech is to transform the speech waveform (which is naturally represented as a convolution) to the frequency domain as a sum of terms which can be separated by ordinary linear fíltering techniques. Techniques for carrying out this procedure are discussed in this chapter, as are several examples of applications of homomorphic speech processing. Chapter 8 deals with the topic of linear predictive coding of speech. This representation is based upon a minimum mean-squared error approximation to the time-varying speech waveform, subject to an assumed linear system model of the speech rignal. This method has been found to be a robust, reliable, and accurate method for representing speech signáis for a wide variety of conditions. The final chapter, Chapter 9, provides a discussion of several speech processing systems in the área of man-machine communication by voice. The purpose of this chapter is twofold: first, to give concrete examples of specific speech processing systems which are used in real worid applications, and second, to show how the ideas developedthroughout the book are applied in representative speech processing systems. The systems discussed in this chapter deal with computer voice response, speaker verification and identification, and speech recognition. The material in this book is intended as a one-semester course in speech process-ing. To aid the teaching process, each chapter (from Chapter 2 to Chapter 8) contatos a set of representative homework problems which are intended to reinforce the ideas discussed in each chapter. Successful completion of a reasonable percentage ofthcse homework problems is essential for a good understanding of the mathematical andtheoretical concepts of speech processing. However, as the reader will see, much of speech processing is, by its very nature, empirical. Thus, some 'hands on' experi-ence is essential to learmng about digital speech processing. In teaching courses based on this book, we have found that a first order approximation to this experience can be obtained by assigning students a term project in one of the following three broadcategories: 1. A literature survey and report 2. A hardware design project 3. A computer project Someguidelines and lists of suggested topics for the three types of projects are given at the end of Chapter 9. We have found that these projects, although demanding, have been popular with our students. We strongly encourage other instructors to incorpórate such projects into courses using this book.