MPEG-2, 4 AAC encoder uses prediction in frequency domain (FDP) to decrease redundancy in the audio signal and ensure more effective coding. For each spectral line, input signals are filtered through the second order adaptive FIR filter called a predictor. Then, instead of processing the signal, the difference between the original signal and the filter output (that is, the prediction error), is passed for further processing. The decoder derives the original signal from the prediction error using a symmetrical block.
You should regularly reset the predictors to their initial state to reduce accumulated calculation error. You may also need to reset the predictors in special cases discussed and described in the ISO-144963 standard. You can reset predictors for the entire spectrum, several scale factor bands, or a selected group of spectral lines.
For more information on the algorithm of filter coefficient adaptation and FDP usage see ISO-144963, clause 6.5.3.2.
To use the FDP prediction tool functions described in this section, follow these steps:
Copyright © 2000 - 2011, Intel Corporation. All rights reserved.