The initial stage of the analysis allows the option of one of a number of different filterbanks to imitate the frequency selectivity of the ear. This division of the binaural signal into a number of frequency bands is included for two main reasons. Firstly, this allows the investigation of the properties of the item under test in different frequency ranges which may be useful in some situations. Secondly, it has been shown that the relationship between the interaural cross-correlation coefficient of a sound and its perceived width is dependent on the audio frequency [Yanagawa et al 1990, Mason et al 2005a]. The separation of the signal into a number of frequency bands allows a frequency-based weighting to be implemented to compensate for this.

The default filterbank that is used in the model is a gammatone filterbank based on the code that was written by Slaney [Slaney 1993], which was subsequently edited by Toshio Irino. The MATLAB script is available as part of an auditory toolbox from the University of Helsinki.

The gammatone filterbank was chosen based on a compromise of accuracy against simplicity. This filterbank gives a reasonable approximation of the frequency and time response of the frequency selectivity part of the auditory system, yet requires relatively little processing power.

Other filterbank options are available for use in the model, including 1/3rd octave and one-octave filters that allow direct comparison with extant measurement techniques. In addition, it is also possible to bypass this stage of the measurement if required, to give a calculation of the characteristics of the entire frequency range of a signal, or for simpler analysis of narrow-band source signals.