Audio Signal Processing and Coding

6.3 FILTER BANKS FOR AUDIO CODING: DESIGN CONSIDERATIONS

This section addresses the issues that govern the selection of a filter bank for audio coding. Efficient coding performance depends heavily on adequately matching the properties of the analysis filter bank to the characteristics of the input signal. Algorithm designers face an important and difficult tradeoff between time and frequency resolution when selecting a filter-bank structure [Bran92a]. Failure to choose a suitable filter bank can result in perceptible artifacts in the output (e.g., pre-echoes) or low coding gain and therefore high bit rates. No single tradeoff between time and frequency resolution is optimal for all signals. We will present three examples to illustrate the challenge facing codec designers. In the first example, we consider the importance of matching time-frequency analysis resolution to the signal-dependent distribution of masking power in the time-frequency plane. The second example illustrates the effect of inadequate frequency resolution on perceptual bit allocation. Finally, the third example illustrates the effect of inadequate time resolution on perceptual bit allocation. These examples clarify the fundamental tradeoff required during filter-bank selection for perceptual coding.

6.3.1 The Role of Time-Frequency Resolution in Masking Power Estimation

Through schematic representations of masking thresholds for castanets and piccolo, Figure 6.4(a,b) illustrates the difficulty of selecting a single ...