Slide 1
I'm Takayuki Shiose from Kyoto University Japan.

Slide 2
At first, I’d like to mention about our background and theoretical basement of our work.

Second, I’ll show you two experimental results and discussions.

Finally, I conclude today’s talk and will present ongoing projects.


Slide 3
At first, I’d like to mention about background of our work.
The blind pedestrians are forced to make a huge effort to travel independently; so they are discouraged from participating in various social events.
I think crossing the street is one of most serious difficulty for the blind.

It is notable fact that some blind people are proficient in navigating for themselves as if they know when they should cross a road.
We call such capability Crossability which is equal to just the obverse side of capability to know the time-to-contact.

Our goal is to identify which acoustic factors affect a pedestrian’s level of Crossablity.

Slide 4
How we should approach it.

The real traffic environment is too complex and too dangerous for us to identify acoustic factors for crossability.
So, we are developing virtual three dimensional acoustic environment. This figure shows the overview of our system.

At first, the car sound sources were recorded in the real world and stored in the Audio Recorder.
Next, main components of this system, a three-dimensional sound space processor (Roland RSS-10) was set.
Finally, processed sounds are mixed, amplified and transferred to listener’s ears.

In ICCHP’04, we presented pre-experiments with sighted subjects and now let me present 2nd report of experiments with both sighted and blind subjects.




Slide 5
We expect a viewpoint of Ecological Psychology to specify what is the key variable for crossability.
You know visual τ is most famous outcome of Ecological Psychology Studies. Here, it is natural that we import this visual τ analogy into acoustic perception. Please, imagine the following scene.

How did you hear? If I’ll explain how I heard it by using onomatopoeia, jobo jobo jobo jobo jobo.





Slide 6
Taking it too easily, we tend to pay attention only to the Doppler effect of direct sounds and to exclude other sounds as noises . 
Certainly, the Doppler effect let us perceive the speed of sound sources. However, the Doppler effect of passing car gives more marked change after the car passed in front of the listener.
We need another factors to perceive crossability before the car reaches in front of them.

You can easily understand how strange this hypothesis is if you see this figure. Listening only direct sounds from the running car equals  paying attention to only retinal expansion rate of the car without other visual targets; asphalt roads, walls, trees, buildings and other pedestrians, that is ambient visual information.

On the other hand, if you stand in the viewpoint of ecological psychology, they insist that we should pay attention rather to ambient sound like reflect sounds and reverberations. So, our main interests are focused on the speed of the sound source and the level of indirect sounds.


Slide 7
Second, let me explain about some experimental results and discussions.

Slide 8
This figure shows the experimental setting. First, the white cane knocking noises was set 8 meters away from the front of the listener.
Second, the car moving sound sources was set to pass about 2 m in front of the listener at a certain speed.

The subjects were instructed to listen to the acoustic events and to click twice, 
clicking first when they noticed the car sound passing over the white cane knocking point,
and second when they heard the car sound passing right in front of them.

Note that sighted subjects were instructed not to watch the screen.


Slide 9
The subjects were tested how correctly they estimated the 8-m distance between the white cane knocking point and the front of them.

Slide 10
13 sighted people and 13 blind people play a role of subjects for our experiments.

Slide 11
The x-axis represents the different car speed conditions from 10 km/h to 40 km/h.
The y-axis represents the time how long the subject perceives the car takes to reach the front of the listener from the white cane knocking point.
The correct time varies in inverse proportion to the speed of the car as described by the red line.

Here, the pink line describes the average of blind pedestrians’ estimation, and the yellow line describes that of sighted people.
You can see that there is similarity between the two graphs and the blind tended to estimate the arrival time longer than the sighted did.

Slide 12
We used a two-way ANOVA to compare means among groups and the Bonferroni/Dunn procedure for Post hoc analyses when the F ratio was significant at p < 0.05. Here, VI is the abbreviation of ``Visual Impairment'' and Sp is that of ``Moving Speed of Sound Source.''

The main effect of the factor ``Speed'' was significant (p<0.01). On the other hand, the main effect of the factor ``Visual Impairment'' was not significant (p=0.1058). The Bonferroni/Dunn test revealed the trend that there were significant differences among different speed conditions at low speed (around 10km/h) and no significant difference at high speed (around 40km/h).

This indicates us that both blind and sighted people tend to underestimate to a greater extent how long a car takes to arrive at the spot the faster it moves.  There was no significant interaction effect, here, although only in cases of blind people, differences between estimated times for different car speed conditions becomes smaller if the car moves faster.

Slide 13
As this figure shows, the blind estimates the time difference to be longer than the sighted.  However, results of the blind and the sighted show same tendency that there are a significant difference among different indirect sound conditions around -40dB and no significant difference near complete reflection (-0dB) and near no indirect sound.

Next, this figure shows the time differences in changing indirect sound conditions. The x-axis represents the different conditions of indirect sound from minus infinite dB to minus 0 dB. The correct time does not vary by dependent on the change of indirect sound conditions as described by the red line.

Here, the pink line describes that of the blind and the yellow one describes that of the sighted.
You can see that there is similarity between the two graphs and the blind tended to estimate the arrival time longer than the sighted did.

Slide 14
We used a two-way ANOVA to compare means among groups and the Bonferroni/Dunn procedure for Post hoc analyses when the F ratio was significant at p < 0.05. Here, IS represents ``Indirect Sound.''

The main effect of the factor ``Indirect Sound'' was significant (p<0.01), as was the main effect of the factor ``Visual Impairment'' (p=0.046). These results suggest that blind people perceive distance to be significantly longer than so sighted people.

Furthermore, the Bonferroni/Dunn test revealed the trend that there were significant differences among different indirect sound conditions around -40dB but no significant difference near the complete reflection condition (-0dB) nor near the condition of no indirect sound condition.

There was no significant interaction effect.

These results indicate that blind subjects estimated or tended to estimate the time-to-contact significantly longer than the sighted subjects did. However, there was no significant difference between the blind and sighted subjects with respect to sensitivity toward each experimental factor, the speed of a moving sound source and the gain level of indirect sounds. 

Slide 15
Finally,  I’ll conclude today’s talk.
These results suggest that we should classify all subjects into groups depending on the difference of high or low sensitivity to changes in speed and indirect sound, not on their sight.  
According to this classification, we should be able to develop training menus that correspond to individual variations of used acoustic clues.  


Slide 16
This figure shows an overview of our ongoing works, Multi speaker version of virtual 3D acoustic training system of crossability. This system can capture the head motion of subjects by using motion tracker FASTRAK. We are going to bring this system to the school of visually impaired this September. So, I think I can report another results until next conference.

That's all.
Thank you.