Audio Signal Processing and Coding

9.3 ANALYSIS/SYNTHESIS AUDIO CODEC (ASAC)

The sinusoidal analysis/synthesis audio codec (ASAC) for robust coding of general audio signals at rates between 6 and 24 kb/s was developed by Edler et al. at the University of Hannover and proposed for MPEG-4 standardization [Edle95] in 1995. An enhanced ASAC proposal later appeared in [Edle96a]. Initially, ASAC segments input audio into analysis frames over which the signal is assumed to be nearly stationary. Sinusoidal synthesis parameters are then extracted according to perceptual criteria, quantized, encoded, and transmitted to the decoder for synthesis. The algorithm distributes synthesis parameters across basic and enhanced bit streams to allow scalable output quality at bitrates of 6 and 24 kb/s. Architecturally, the ASAC scheme (Figure 9.3) consists of a pre-analysis block for window selection and envelope extraction, a sinusoidal analysis-by-synthesis parameter estimation block, a perceptual model, and a quantization and coding block. Although it bears similarities to sinusoidal speech coding [McAu86] [Geor92] [Geor97] and music synthesis [Serr90] algorithms that have been available for some time, the ASAC coder also incorporates some new techniques. In particular, whereas previous speech-specific sinusoidal coders emphasized wave-form matching by minimizing reconstruction error norms such as the mean square error (MSE), ASAC disregards classical error minimization criteria and instead selects sinusoids in decreasing order of ...