Audio Signal Processing and Coding

9.4 HARMONIC AND INDIVIDUAL LINES PLUS NOISE CODER (HILN)

The ASAC algorithm outperformed speech-specific algorithms at the same bit rate in subjective tests for certain test signals, particularly spectrally complex music characterized by large numbers of nonharmonically related sinusoids. The original ASAC, however, failed to match speech codec performance for other test signals such as clean speech. As a result, the ASAC core was embedded in an enhanced algorithm [Purn98] intended to better match the coder's signal model with diverse input signal characteristics. In research proposed as part of an MPEG-4 “core experiment” [Purn97], Purnhagen et al. developed an “object-based” algorithm. In this approach, harmonic sinusoid, individual sinusoid, and colored noise objects could be combined in a hybrid source model to create a parametric signal representation. The enhanced algorithm, known as the “harmonic and individual lines plus noise” (HILN) is architecturally very similar to the original ASAC, with some modifications.

Like ASAC, the HILN coder (Figure 9.4) segments the input signal into 32-ms overlapping analysis frames and extracts a parametric temporal envelope during preanalysis. Unlike ASAC, however, the iterative analysis-synthesis block is extended to include a cascade of analysis stages for each of the available object types. In the enhanced analysis-synthesis system of HILN, harmonic analysis is applied first, followed by individual spectral line analysis, followed by ...