I already responded to this but here's what Holman has to say about it:
"There can be a case made for capturing high-resolution
data, and employing different smoothing schemes in
separate frequency ranges to present the data. The
frequency range of the lowest critical band of hearing is
quite wide, whereas we definitely hear the note-by-note
variations in level within the band caused by the
interaction of the source and receive location with the
room acoustics, in particular caused by low-frequency
room modes. Thus a critical band filter is too broad to
explain what is heard in this case. So in one way of
looking at the problem, we might use fractional
bandwidth filters of high resolution, say 1/24th octave,
at low frequencies, below the Schroeder frequency, and
cross over to ERB above the Schroeder frequency."
From AES paper "First Results from a Large-Scale Measurement Program for Home Theaters"