Audio Signal Processing and Coding

9.1 INTRODUCTION

This chapter addresses perceptual coding algorithms based on sinusoidal models. Although sinusoidal signal models have been applied successfully since the 1980s in speech coding [Hede81] [Alme83] [McAu86] [Geor87] [Geor92] and music synthesis [Serr90], perceptual properties were not introduced in sinusoidal modeling until later [Edle96c] [Pena96] [Levin98a] [Pain01]. The advent of MPEG-4 standardization established new research goals for high-quality coding of general audio signals at bit rates in the range of 6–24 kb/s. In experiments reported as part of the MPEG-4 standardization effort, it was determined that sinusoidal coding is capable of achieving good quality at low rates without being constrained by a restrictive source model. Furthermore, unlike CELP and other classical low rate speech coding models, the parametric sinusoidal coding is amenable to pitch and time-scale modification at the decoder. Additionally, the emergence of Internet-based streaming audio has motivated considerable research on the application of sinusoidal signal models to high-quality audio coding at low bit rates. For example, Levine and Smith developed a hybrid sinusoidal-filter-bank coding scheme that achieves very high quality at rates around 32 kb/s [Levin98a] [Levi99].

This chapter describes some of the sinusoidal algorithms for low rate audio coding that exploit perceptual properties. In Section 9.2, we review the classical sinusoidal model. Section 9.3 presents the analysis/synthesis ...