In summary, we noted above the following general guidelines
regarding series vs. parallel elementary-section implementations:
Series sections are preferred when all sections contribute to the
same passband, such as in a lowpass, highpass, bandpass, or bandstop
filter.
Parallel sections are usually preferred when
the sections have disjoint passbands, such as a formant filter bank
used in voice models. Another example would be the phase vocoder
filter bank [22].