Audio Signal Processing and Coding

7.1 INTRODUCTION

Transform coders make use of unitary transforms (e.g., DFT, DCT, etc.) for the time/frequency analysis section of the audio coder shown in Figure 1.1. Many transform coding schemes for wideband and high-fidelity audio have been proposed, starting with some of the earliest perceptual audio codecs. For example, in the mid-1980s, Krahe applied psychoacoustic bit allocation principles to a transform coding scheme [Krah85] [Krah88]. Schroeder [Schr86] later extended these ideas into multiple adaptive spectral audio coding (MSC). The MSC utilizes a 1024-point DFT, then groups coefficients into 26 subbands, inspired by the critical bands of the ear. This chapter gives overview of algorithms that were proposed for transform coding of high-fidelity audio following the early work of Schroeder [Schr86].

The Chapter is organized as follows. Sections 7.2 through 7.5 describe in some detail the transform coding algorithms proposed by Brandenburg, Johnston, and Mahieux [Bran87b] [John88a] [Mahi89] [Bran90]. Most of this research became connected with the MPEG standardization, and the ISO/IEC eventually clustered these algorithms into a single candidate algorithm called adaptive spectral entropy coding (ASPEC) [Bran91] of high quality music signals. The ASPEC algorithm (Section 7.6) has become part of the ISO/IEC MPEG-1 [ISOI92] and the MPEG-2/BC-LSF [ISOI94a] audio coding standards. Sections 7.7 and 7.8 are concerned with two transform coefficient substitution schemes, namely ...