7.1. Psychoacoustic modelling
167
Figure 7.2
In-band masking level derived and plotted (bottom) from an example spectrum (top)
for 15 critical bands of constant Bark width.
7.1.7
Other critical-band spreading functions
Experimental results indicate that sound in one critical band will spread its influence
into neighbouring critical bands. Given a loud sound in a particular band, depending
upon how far away from the band centre we look, this will have an effect upon other
nearby sounds. Many models exist to quantify this spreading for a particular sound,
known variously as critical-band spreading function, noise masking curve and lateral
inhibition function.
Several approximation functions were empirically derived by various authors, and are
plotted in Figure 7.3. Of the curves shown, only the model of Cheng and O’Shaughnessy
[5,6] attempts to account for lateral inhibition [3] where a sound in one band de-sensitises
neighbouring frequencies. Jayant et al. [7], Virag [8] and Sen [9] report very simi-
lar curves, differing mainly in the upper frequency region. However Sen introduces
corrections for both absolute frequency (the plot shows frequency relative to critical
band centre frequency) and absolute power, in an attempt to improve accuracy. For
the plot shown, a centre frequency of 1 kHz and absolute power of 70 dB
SPL
were assumed. The former two authors do not account for absolute position in frequency
or amplitude.
The curve used by Hermansky [4] has a flat-topped response that not only approxi-
mates the bandwidth of each critical band, but accounts for spreading with a depend-
ence on absolute frequency and amplitude in a fashion similar to the model from Sen
[9]. Most importantly the flat top of the function eases the computational load of the
model.
168
Do'stlaringiz bilan baham: |