Modeling Auditory Localization of Subwoofer Signals in Multi-Channel Loudspeaker Arrays

1
Simulating the precedence effect by means of
autocorrelation
Jonas Braasch Architectural Acoustics
Group Communication Acoustics and Aural
Architecture Research Laboratory
(CA3RL) Rensselaer Polytechnic Institute, Troy,
NY http//symphony.arch.rpi.edu/carl
2
Summing localization region
Localization Dominance region
Region with two auditory events
lead
lag
or interstimulus interval (ISI)
(Blauert Spatial Hearing, 1983)
The precedence effect
3
Objectives

1. Model should be able to focus on stationary
   (non-onset) cues
2. Analytical optimization of inhibition parameters
3. Accurate simulation of data with ILD cues

4
Time course
Unusual approach using a monophonic lead/lag pair
5
Road Map

1. use autocorrelation to estimate lag parameters
   (lag delay and amplitude)
2. use a filter to eliminate lag from total signal
3. extent mechanism to comply with the concept of
   auditory bands
4. integrate algorithm in binaural model (e.g.,
   eliminate lag in each channel)
5. evaluate model performance using literature data

6
a
t
Autocorrelation function (ACF) for a lead/lag
pair (one channel) Stimulus 100 ms/broadband
noise burst (20-2000 Hz) Interstimulus Interval
5 ms.
7
Lead/lag autocorrelation function
8
autocorrelation function decomposition
9
relative lag amplitude
A
B
t0
tD
10
lag removal filter
ACF model
classic inhibition approach
excitation
Parameters not adapted to stimuli
tD
tinh
aainh
aalag/alead
inhibition (no excitation)
inhibition
Filter Impulse Response
11
autocorrelation evaluation of filtered signal
lead only
Leadlag
Lead, after lag removal
error
12
auditory filter simulation
13
broadband AFC resynthesis
Orthogonal for separate freqs
14
cross-correlation analysis for binaural signal
15
binaural model architecture
16
model data
psychoacoustic data
lead position
lag position
ISIInterstimulus Interval
17
exp of Colburn Dizon (2006)
18
(No Transcript)
19
Conclusions

1. Model processes sound based on stationary
   (non-onset) cues and accurate simulates of data
   with ILD cues
2. model optimizes inhibition parameters
   analytically

20
Lindemann model Peak is off-center at position of
the lead
Simple Cross-correlation model Peak is on-center
between the positions of lead and lag
Lindemann Model (1985)
21
IIR filter to remove lag from lead/lag pair.
ACF before filtering
influence of lag
ACF after filtering
no lag