Audio Signal Processing and Coding

7.4 BRANDENBURG-JOHNSTON HYBRID CODER

Johnston and Brandenburg [Bran90] collaborated in 1990 to produce a hybrid coder that, strictly speaking, is both a subband and transform coding algorithm. The idea behind the hybrid coder is to improve time and frequency resolution relative to OCF and PXFM by constructing a filter bank that more closely resembles the auditory filter bank. This is accomplished at the encoder by first splitting the input signal into four octave-width subbands using a QMF filter bank.

The decimated output sequence from each subband is then followed by one or more transforms to achieve the desired time/frequency resolution, Figure 7.3(a). Both the DFT and the MDCT were investigated. Given the tiling of the time-frequency plane shown in Figure 7.3(b), frequency resolution at low frequencies (23.4 Hz) is well matched to the ear, while the time resolution at high frequencies (2.7 ms) is sufficient for pre-echo control.

The quantization and coding schemes of the hybrid coder combine elements from both PXFM and OCF. Masking thresholds are estimated using the PXFM approach for eight time slices in each frequency subband. A more sophisticated tonality estimate was defined to replace the SFM (Eq. (5.13)) used in PXFM, however, such that tonality is estimated in the hybrid coder as a local characteristic of each individual spectral line. Predictability of magnitude and phase spectral components across time is used to evaluate tonality instead of just global spectral shape ...