dspace.jaist.ac.jp...Japan Advanced Institute of Science and Technology JAIST Repository Title...
Transcript of dspace.jaist.ac.jp...Japan Advanced Institute of Science and Technology JAIST Repository Title...
![Page 1: dspace.jaist.ac.jp...Japan Advanced Institute of Science and Technology JAIST Repository Title 聴覚生理モデルを用いた音源方向定位に関する基礎的 研究 Author(s)](https://reader031.fdocument.pub/reader031/viewer/2022011822/5ecc54088036652cbb38fd2d/html5/thumbnails/1.jpg)
Japan Advanced Institute of Science and Technology
JAIST Repositoryhttps://dspace.jaist.ac.jp/
Title聴覚生理モデルを用いた音源方向定位に関する基礎的
研究
Author(s) 伊藤, 一仁
Citation
Issue Date 1998-03
Type Thesis or Dissertation
Text version author
URL http://hdl.handle.net/10119/1133
Rights
Description Supervisor:赤木 正人, 情報科学研究科, 修士
![Page 2: dspace.jaist.ac.jp...Japan Advanced Institute of Science and Technology JAIST Repository Title 聴覚生理モデルを用いた音源方向定位に関する基礎的 研究 Author(s)](https://reader031.fdocument.pub/reader031/viewer/2022011822/5ecc54088036652cbb38fd2d/html5/thumbnails/2.jpg)
Study on Sound Localization
using Auditory Function Model
Kazuhito Ito
School of Information Science,
Japan Advanced Institute of Science and Technology
February 13, 1998
Keywords: ITD,Je�ress's model,impulse,synaptic transmission,inhibition.
Sound localization is an auditory function to detect of sound source location using
the interaural time di�erence(ITD) and the interaural level di�erence(ILD) between ears
at which the sound waves arrive. In this study, a function model of the auditory sound
localization based on the interaural time di�erence(ITD) is presented.
Sound waves arriving at ears are decomposed into their frequency components and are
changed into impulse trains by the auditory periphery. The impulse trains keep timing
information in itselves because auditory nerves tend to �re at a certain phase of stimuli.
The system for detecting ITDs exists in medial superior olive(MSO), where the nerves
coming from left and right ears cross each other. The Je�ress model is one of model
circuits for the detection. This model has been approved because of simple theory and
structual analogy between the model and tissues in MSO. At the beginning in this study,
the Je�ress model is implemented computationally to examine its mechanism and to
improve.
The Je�ress model is represented as a circuit which consists of some coincidence de-
tectors and two nerve �bers from left and right ears. The detectors �re only when impulse
trains coming from both sides through nerve �bers arrive simultaneously. Thus the model
can caliculate ITDs with correlation between impulse trains coming from both sides(Figure
1).
When a sound source is placed in front of the head, arrival times are equal on the left
and right pathways, because the times that sound wave comes to ears and impulse trains
come to the circuit are equal. Then, the middle detector in the circuit responds most
strongly. The position of the responding detector varies as a sound source moves.
Auditory nerve �bers do not always �re at a certain phase of stimuli. Impulses uctuate
temporally and it is di�cult to detect ITDs with the correlation between those impulses.
Copyright c 1998 by Kazuhito Ito
1
![Page 3: dspace.jaist.ac.jp...Japan Advanced Institute of Science and Technology JAIST Repository Title 聴覚生理モデルを用いた音源方向定位に関する基礎的 研究 Author(s)](https://reader031.fdocument.pub/reader031/viewer/2022011822/5ecc54088036652cbb38fd2d/html5/thumbnails/3.jpg)
sound source
left ear right ear
relay station in brain
sound source
delay lineleft ear right ear
coincidence detector
relay station in brain
delay line
coincidence detector
Figure 1: Je�ress model
Figure 2 shows the period histogram of impulse trains with large uctuation and the
output from the circuit with correlation between impulses. The ITD is not clear.
0 50 100 150 200 250 300 3500
50
100
150
200
250
300
f=3kHz
PHASE [o]
SP
IKE
S
−600 −400 −200 0 200 400 6000
100
200
300
400
ITD [µs]
coin
cide
nces
f=3kHz
*
Figure 2: period histogram and the output
Therefore, in this study, the time length and the amplitude of impulses in the circuit
are modi�ed into a reasonable shape like a sawtooth wave, to reduce errors of the detection
against uctuation of impulse trains. Figure 3 shows the histogram of the same impulse
trains as Figure 2 and the output from the circuit with sawtooth waves. The peak shows
the correct ITD clearly.
The model has another problem, related to phase ambiguity that detectors in the
circuit respond to more than one ITD, as represented �t + nT (n is an integer. T is a
period.) when the actual ITD is �t. To solve the problem of phase ambiguity, the circuits
2
![Page 4: dspace.jaist.ac.jp...Japan Advanced Institute of Science and Technology JAIST Repository Title 聴覚生理モデルを用いた音源方向定位に関する基礎的 研究 Author(s)](https://reader031.fdocument.pub/reader031/viewer/2022011822/5ecc54088036652cbb38fd2d/html5/thumbnails/4.jpg)
0 50 100 150 200 250 300 3500
50
100
150
200
250
300
f=3kHz
PHASE [o]
SP
IKE
S
−600 −400 −200 0 200 400 6000
100
200
300
400
ITD [µs]
coin
cide
nces
f=3kHz*
Figure 3: period histogram and the output
are arranged in each frequincy band and the detectors in the circuits are grouped with
the same ITD. The most �ring group indicates the actual ITD(Figure 4). In fact, it is
found that ITD detectors in organisms are arranged systematically along frequency axis.
The model can be improved with the physiological knowledge. However, the simple
mechanism to detect the correlation of stimuli from ears, like this model, can not work
well under real environments with noise from some sources, like orgnisms do.
To investigate these issues in more detail, signal representations in orgnisms, such
as nervous impulse or synaptic transmission, are modeled computationally according to
physiological knowledge. For example, an impulse is a deviation of membrane potential
called 'action potential' with certain time duration. Additionally, synaptic transmissions
extend the width of action potentials. Then, timing information may become obscure.
These signals are applied to the coincidence detector circuits of ITD in the model. The
results of the simulation using a coincidence detector circuit show that �rings of one
coincidence detection spread over the circuits, in response to just one correlation (Figure
5). Thus, it is di�cult to determine the actual ITD using one coincidence detector circuit.
To determine ITD with more accuracy and calculate it more e�ciently, selecting by
many thresholds and inhibition are assumed. Since the coincidence detector indicating
the actual ITD tends to �re earlier than others, the �rst �ring event at the actual ITD
excits its own postsynaptic neurons and inhibits other ones(Figure 6). Consequently, the
model with inhibition can improve accuracy to detect ITDs(Figure 7).
In conclusion, nervous impulses and synaptic transmission from left and right ears
in MSO are modeled computationally and are applied to a past coincidence detecotor
circuit model to detect ITD. The results show that the past model is not able to detect
actual ITD independently. Then, the selection by many thresholds and the inhibition are
modeled to improve detection accuracy of ITD.
3
![Page 5: dspace.jaist.ac.jp...Japan Advanced Institute of Science and Technology JAIST Repository Title 聴覚生理モデルを用いた音源方向定位に関する基礎的 研究 Author(s)](https://reader031.fdocument.pub/reader031/viewer/2022011822/5ecc54088036652cbb38fd2d/html5/thumbnails/5.jpg)
−600 −400 −200 0 200 400 6000
0.5
1
RA
TE
−600 −400 −200 0 200 400 6000
200
400
1.5k
Hz
−600 −400 −200 0 200 400 6000
200
400
3kH
z
−600 −400 −200 0 200 400 6000
200
400
4kH
z
−600 −400 −200 0 200 400 6000
200
400
5kH
z
ITD [µs]
Figure 4: solution for phase ambiguity
01
23
4
−2−1
01
2−80
−75
−70
−65
−60
−55
−50
−45
−40
−35
−30
spatial summation
TIME[ms]
broad firing
AZIMUTH(ITD)[ms]
corr
elat
ion:
EP
SP
[mV
]
Figure 5: spatial summation in the circuits
4
![Page 6: dspace.jaist.ac.jp...Japan Advanced Institute of Science and Technology JAIST Repository Title 聴覚生理モデルを用いた音源方向定位に関する基礎的 研究 Author(s)](https://reader031.fdocument.pub/reader031/viewer/2022011822/5ecc54088036652cbb38fd2d/html5/thumbnails/6.jpg)
0.7 0.8 0.9 1 1.1
−0.50
0.5−41
−40.5
−40
−39.5
−39
TIME[ms]
EPSP
peak
AZIMUTH(ITD)[ms]
threshold
EP
SP
[mV
]
0.7 0.8 0.91 1.1
−0.5
0
0.50
0.5
1
TIME[ms]AZIMUTH(ITD)[ms]
Timing of FiringE
PS
P[m
V]
Figure 6: the di�erence in the timing of �ring
0
5
10
15
20
−1
−0.5
0
0.5
1−80
−70
−60
−50
−40
−30
with inhibition
sum of 3 types
TIME[ms]AZIMUTH(ITD) [ms]
PS
P w
/inhi
bitio
n [m
s]
Figure 7: psp with inhibition
5