Precedence effect (precedenceProc.m
)
The precedence effect describes the ability of humans to fuse and localize the
sound based on the first-arriving parts, in the presence of its successive
version with a time delay below an echo-generating threshold [Wallach1949]. The
effect of the later-arriving sound is suppressed by the first part in the
localization process. The precedence effect processor in Auditory front-end models this, with
the strategy based on the work of [Braasch2013]. The processor detects and
removes the lag from a binaural input signal with a delayed repetition, by means
of an autocorrelation mechanism and deconvolution. Then it derives the ITD and
ILD based on these lag-removed signals.
The input to the precedence effect processor is a binaural time-frequency signal
chunk from the gammatone filterbank. Then for each chunk a pair of ITD and
ILD values is calculated as the output, by integrating the ITDs and ILDs
across the frequency channels according to the weighted-image model
[Stern1988], and through amplitude-weighted summation. Since these ITD/ILD
calculation methods of the precedence effect processor are different from what
are used for the Auditory front-end ITD and ILD processors, the Auditory front-end ITD and ILD
processors are not connected to the precedence effect processor. Instead the
steps for the correlation analyses and the ITD/ILD calculation are coded
inside the processor as its own specific techniques. Table 34
lists the parameters needed to operate the precedence effect processor.
Table 34 List of parameters related to the auditory representation
’precedence’
.
Parameter |
Default |
Description |
prec_wSizeSec |
20E-3 |
Window duration in s |
prec_hSizeSec |
10E-3 |
Window step size in s |
prec_maxDelaySec |
10E-3 |
Maximum delay in s for autocorrelation computation |
Fig. 42 shows the output from a demonstration script
DEMO_precedence.m
. The input signal is a 800-Hz wide bandpass noise of 400
ms length, centered at 500 Hz, mixed with a reflection that has a 2 ms delay,
and made binaural with an ITD of 0.4 ms and a 0-dB ILD. During the
processing, windowed chunks are used as the input, with the length of 20 ms. It
can be seen that after some initial confusion, the processor estimates the
intended ITD and ILD values as more chunks are analyzed.
[Braasch2013] | Braasch, J. (2013), “A precedence effect model to simulate localization
dominance using an adaptive, stimulus parameter-based inhibition process.”
The Journal of the Acoustical Society of America 134(1), pp. 420–35. |