1
00:00:00,240 --> 00:00:03,000
okay so first of all hello everyone I'm

2
00:00:03,000 --> 00:00:05,640
Raz from Israel I'm security is such a

3
00:00:05,640 --> 00:00:08,099
disabled Labs of Bangor University

4
00:00:08,099 --> 00:00:10,139
and today we'll present a little silver

5
00:00:10,139 --> 00:00:12,300
it's an optical sound recovery from

6
00:00:12,300 --> 00:00:14,280
light with reflecting objects

7
00:00:14,280 --> 00:00:16,139
and I'm really thankful for for being

8
00:00:16,139 --> 00:00:19,199
here and I'm excited so let's begin

9
00:00:19,199 --> 00:00:21,240
and so first of all this talk is based

10
00:00:21,240 --> 00:00:22,920
on a research that was covered by Dr

11
00:00:22,920 --> 00:00:26,000
Benassi professors

12
00:00:26,000 --> 00:00:28,880
and myself

13
00:00:28,880 --> 00:00:30,660
agenda for today

14
00:00:30,660 --> 00:00:32,520
and first of all we start with some

15
00:00:32,520 --> 00:00:34,200
introduction and we'll review we will

16
00:00:34,200 --> 00:00:36,239
review the related look for this paper

17
00:00:36,239 --> 00:00:38,040
then we will see the third model of the

18
00:00:38,040 --> 00:00:39,059
attack and we'll discuss some

19
00:00:39,059 --> 00:00:40,559
interesting experiments we did during

20
00:00:40,559 --> 00:00:41,820
the analysis

21
00:00:41,820 --> 00:00:43,920
then we will discuss on how to recover

22
00:00:43,920 --> 00:00:45,960
speech from Optical measurements and we

23
00:00:45,960 --> 00:00:48,059
will hear some nice speech recoveries

24
00:00:48,059 --> 00:00:49,739
and we will review the takeaways of the

25
00:00:49,739 --> 00:00:51,300
talk hopefully we'll have some time for

26
00:00:51,300 --> 00:00:53,640
Q a

27
00:00:53,640 --> 00:00:55,620
so first of all what do you know

28
00:00:55,620 --> 00:00:57,059
regarding the picture presented on the

29
00:00:57,059 --> 00:00:58,739
right side of the slide

30
00:00:58,739 --> 00:01:00,899
and there are some few interesting facts

31
00:01:00,899 --> 00:01:03,539
about it the first fact is that as you

32
00:01:03,539 --> 00:01:05,339
can see it's a picture of an eagle

33
00:01:05,339 --> 00:01:07,380
but the more interesting fact is that it

34
00:01:07,380 --> 00:01:09,180
was given as a gift to the United States

35
00:01:09,180 --> 00:01:11,520
ambassador to the Soviet Union from

36
00:01:11,520 --> 00:01:14,460
Soviet Union in 1945.

37
00:01:14,460 --> 00:01:16,680
but if you really if you review it

38
00:01:16,680 --> 00:01:18,900
internally it wasn't a relative it was

39
00:01:18,900 --> 00:01:21,659
actually the first last New Years

40
00:01:21,659 --> 00:01:24,180
yeah so it seems there is some delay we

41
00:01:24,180 --> 00:01:27,119
see your slides in time uh what is that

42
00:01:27,119 --> 00:01:28,080
one

43
00:01:28,080 --> 00:01:31,399
that monitor is the

44
00:01:32,400 --> 00:01:34,560
what's like do you see now I see it yes

45
00:01:34,560 --> 00:01:38,240
yeah how can we fix this

46
00:01:39,180 --> 00:01:42,840
is it because we see this live correctly

47
00:01:42,840 --> 00:01:44,700
which means the signal is arriving in

48
00:01:44,700 --> 00:01:46,259
this room correctly

49
00:01:46,259 --> 00:01:48,240
exactly so

50
00:01:48,240 --> 00:01:50,159
what

51
00:01:50,159 --> 00:01:53,540
which side is it now

52
00:01:54,659 --> 00:01:57,119
yeah I think another way since we are

53
00:01:57,119 --> 00:01:57,899
actually

54
00:01:57,899 --> 00:02:00,840
yeah I think we're on time exactly so go

55
00:02:00,840 --> 00:02:02,520
ahead maybe click on the next light and

56
00:02:02,520 --> 00:02:05,060
then wait five seconds

57
00:02:05,060 --> 00:02:07,140
in the meantime I try to find someone

58
00:02:07,140 --> 00:02:08,459
for

59
00:02:08,459 --> 00:02:11,300
Okay so

60
00:02:11,819 --> 00:02:13,680
um so as I said if you review it

61
00:02:13,680 --> 00:02:15,360
internally it was actually the first

62
00:02:15,360 --> 00:02:17,340
covered listening device that utilized

63
00:02:17,340 --> 00:02:19,920
passive techniques it contains a passive

64
00:02:19,920 --> 00:02:21,360
device which is considered the first

65
00:02:21,360 --> 00:02:23,099
predecessor of firefighter technology

66
00:02:23,099 --> 00:02:27,000
which was used for review topping

67
00:02:27,000 --> 00:02:29,160
you may know this device by the name the

68
00:02:29,160 --> 00:02:32,540
great deal plug or the thing

69
00:02:33,599 --> 00:02:35,940
now I want to ask you another question

70
00:02:35,940 --> 00:02:37,620
take a look at the picture on the right

71
00:02:37,620 --> 00:02:39,000
side of the screen

72
00:02:39,000 --> 00:02:40,620
you can see there are a few unaltered

73
00:02:40,620 --> 00:02:42,180
reflective lightweight objects which we

74
00:02:42,180 --> 00:02:43,920
call the Little Seal box

75
00:02:43,920 --> 00:02:45,599
do you think that attackers can use

76
00:02:45,599 --> 00:02:47,400
these objects as Optical implants for

77
00:02:47,400 --> 00:02:50,780
the purpose of speeches dropping

78
00:02:51,800 --> 00:02:54,120
so the short answer for this question

79
00:02:54,120 --> 00:02:56,220
question is yes and I'm going to show

80
00:02:56,220 --> 00:02:59,180
you how to work this talk

81
00:02:59,400 --> 00:03:01,560
let's discuss some related eavesdropping

82
00:03:01,560 --> 00:03:05,060
research resources researchers

83
00:03:05,580 --> 00:03:07,560
so first of all in recent news the

84
00:03:07,560 --> 00:03:09,239
scientific Community has offered various

85
00:03:09,239 --> 00:03:11,400
ways to recover sound and these ways are

86
00:03:11,400 --> 00:03:13,260
divided into two main categories

87
00:03:13,260 --> 00:03:15,000
and the internal methods and the

88
00:03:15,000 --> 00:03:17,159
external methods

89
00:03:17,159 --> 00:03:19,140
internal methods or methods that obtain

90
00:03:19,140 --> 00:03:21,540
data data relying on a device located in

91
00:03:21,540 --> 00:03:23,400
a proximity to the victim

92
00:03:23,400 --> 00:03:25,620
for example it was demonstrated how data

93
00:03:25,620 --> 00:03:27,720
obtained from a motion sensor or a

94
00:03:27,720 --> 00:03:30,360
magnetic data obtained from a hard drive

95
00:03:30,360 --> 00:03:33,300
can can be used to recover speech

96
00:03:33,300 --> 00:03:35,459
on the other hand the other category is

97
00:03:35,459 --> 00:03:37,620
external methods which are methods that

98
00:03:37,620 --> 00:03:40,260
rely on data obtained by a device that

99
00:03:40,260 --> 00:03:42,900
is not located in near the victim the

100
00:03:42,900 --> 00:03:45,000
primary methods under this topic is the

101
00:03:45,000 --> 00:03:47,040
laser microphone which will utilize the

102
00:03:47,040 --> 00:03:48,780
laser transceiver to recover speech

103
00:03:48,780 --> 00:03:50,879
visual microphone which uses a high

104
00:03:50,879 --> 00:03:53,099
frequency video camera to analyze the

105
00:03:53,099 --> 00:03:55,680
vibration of objects and recover speech

106
00:03:55,680 --> 00:03:58,379
from them for example a bag of chips

107
00:03:58,379 --> 00:04:00,360
then the last method in this category is

108
00:04:00,360 --> 00:04:02,159
landform which utilize a photodel to

109
00:04:02,159 --> 00:04:04,080
recover speech by analyzing light bulbs

110
00:04:04,080 --> 00:04:06,739
of vibration

111
00:04:06,900 --> 00:04:09,000
in general if I have to summarize this

112
00:04:09,000 --> 00:04:10,860
section of related work each of the

113
00:04:10,860 --> 00:04:12,239
methods I presented to you today is

114
00:04:12,239 --> 00:04:13,680
limited by at least one of the following

115
00:04:13,680 --> 00:04:14,939
limitations

116
00:04:14,939 --> 00:04:16,978
first it might rely on a compromised

117
00:04:16,978 --> 00:04:19,380
device it might be active which makes it

118
00:04:19,380 --> 00:04:21,779
much easier to detect it it can be

119
00:04:21,779 --> 00:04:23,880
applied in real time or it relies on an

120
00:04:23,880 --> 00:04:26,820
uncommon object in the victim's room

121
00:04:26,820 --> 00:04:28,800
now let's discuss the 10 model of the

122
00:04:28,800 --> 00:04:31,320
little silver attack

123
00:04:31,320 --> 00:04:33,360
first of all we assume that the victim

124
00:04:33,360 --> 00:04:35,160
makes a call or attend the meeting from

125
00:04:35,160 --> 00:04:36,840
an office or a room that contains a

126
00:04:36,840 --> 00:04:38,639
little silver in the form of a light

127
00:04:38,639 --> 00:04:40,020
which any object

128
00:04:40,020 --> 00:04:42,180
second we will assume that the sound

129
00:04:42,180 --> 00:04:43,800
from the victim's conversation create

130
00:04:43,800 --> 00:04:45,540
fluctuations on the surfaces of the

131
00:04:45,540 --> 00:04:47,699
light reflective object for example a

132
00:04:47,699 --> 00:04:49,440
nice coffee can or a desk element that

133
00:04:49,440 --> 00:04:52,199
is placed on the desk

134
00:04:52,199 --> 00:04:54,360
in addition we will assume that the evil

135
00:04:54,360 --> 00:04:56,220
detects the photo diode and the light to

136
00:04:56,220 --> 00:04:58,320
reach any object via telescope

137
00:04:58,320 --> 00:05:00,360
the optical signal from the photodiode

138
00:05:00,360 --> 00:05:02,040
is sampled via an analog to digital

139
00:05:02,040 --> 00:05:04,380
converter and after that another view to

140
00:05:04,380 --> 00:05:07,440
recover the acoustic signal

141
00:05:07,440 --> 00:05:09,900
foreign perspective this technology is

142
00:05:09,900 --> 00:05:12,180
external it's passive and it can be

143
00:05:12,180 --> 00:05:14,340
applied in real time it is also based on

144
00:05:14,340 --> 00:05:16,080
really common object that can be used in

145
00:05:16,080 --> 00:05:18,120
the room

146
00:05:18,120 --> 00:05:19,860
now let's analyze the physical

147
00:05:19,860 --> 00:05:22,460
phenomenon

148
00:05:22,680 --> 00:05:24,900
so the first experiment we did is to

149
00:05:24,900 --> 00:05:26,400
measure the vibration of reflective

150
00:05:26,400 --> 00:05:29,400
objects we took a 50 grams anyway and

151
00:05:29,400 --> 00:05:32,100
connected it to a gyroscope then we play

152
00:05:32,100 --> 00:05:33,660
the frequency scan from a speaker that

153
00:05:33,660 --> 00:05:35,100
was placed 10 centimeters from the

154
00:05:35,100 --> 00:05:35,820
weight

155
00:05:35,820 --> 00:05:39,539
at 75 decibels which is important

156
00:05:39,539 --> 00:05:41,880
there is also actually satisfying we can

157
00:05:41,880 --> 00:05:43,800
see in the right in the figure in the

158
00:05:43,800 --> 00:05:46,139
right of in right side of the slide that

159
00:05:46,139 --> 00:05:48,419
the weight vibrates based on the nearby

160
00:05:48,419 --> 00:05:51,240
sound as can be said and that's what we

161
00:05:51,240 --> 00:05:52,919
expected and we go on to the next

162
00:05:52,919 --> 00:05:55,340
experiment

163
00:05:55,680 --> 00:05:57,600
next we wanted to analyze the light

164
00:05:57,600 --> 00:05:59,699
reflected from the work we use the

165
00:05:59,699 --> 00:06:01,560
telescope directed on the weight and we

166
00:06:01,560 --> 00:06:03,780
monitor the photodiode on it then we

167
00:06:03,780 --> 00:06:05,639
sample the photodiode with an ADC and

168
00:06:05,639 --> 00:06:07,020
play the frequency scan in the same

169
00:06:07,020 --> 00:06:09,240
settings as in the previous experiments

170
00:06:09,240 --> 00:06:10,979
we obtain the optical signals from the

171
00:06:10,979 --> 00:06:12,900
photo there three times the first time

172
00:06:12,900 --> 00:06:14,400
was when the light in the room were on

173
00:06:14,400 --> 00:06:17,460
and the way it was uh regularly the

174
00:06:17,460 --> 00:06:18,840
second time was when the weight was

175
00:06:18,840 --> 00:06:20,580
covered with with black tape so it

176
00:06:20,580 --> 00:06:22,740
wouldn't deflect light and the third

177
00:06:22,740 --> 00:06:24,180
time was when the light in the room were

178
00:06:24,180 --> 00:06:26,360
off

179
00:06:26,759 --> 00:06:29,160
there is also there is also as expected

180
00:06:29,160 --> 00:06:30,600
we saw that

181
00:06:30,600 --> 00:06:32,400
um after calculating this enough on the

182
00:06:32,400 --> 00:06:34,380
optical measurements we saw that nearby

183
00:06:34,380 --> 00:06:36,180
sounds can be recovered only when lights

184
00:06:36,180 --> 00:06:38,759
are on and object reflects light which

185
00:06:38,759 --> 00:06:40,440
means the risk of spiritual cover is

186
00:06:40,440 --> 00:06:43,620
only affected the risk of spirit

187
00:06:43,620 --> 00:06:46,199
recovery is only exists unexisted when

188
00:06:46,199 --> 00:06:49,919
the lights reflects from the object

189
00:06:49,919 --> 00:06:52,319
now let's discuss on how to isolate the

190
00:06:52,319 --> 00:06:55,500
sound recovery from the optical signal

191
00:06:55,500 --> 00:06:57,539
so first after we get the whole signal

192
00:06:57,539 --> 00:06:58,979
we follow the we'll follow the next

193
00:06:58,979 --> 00:07:00,120
steps

194
00:07:00,120 --> 00:07:01,979
we will first filter the side effects

195
00:07:01,979 --> 00:07:04,500
that are in the result of sound and for

196
00:07:04,500 --> 00:07:06,479
example side effect that can be out of

197
00:07:06,479 --> 00:07:09,780
the electricity harmonics of 100 Hertz

198
00:07:09,780 --> 00:07:12,240
to filter them we use one we use that we

199
00:07:12,240 --> 00:07:14,580
use the bands to fill those

200
00:07:14,580 --> 00:07:16,560
then we want to enhance the speech by

201
00:07:16,560 --> 00:07:18,419
normalizing the virus of the signal to

202
00:07:18,419 --> 00:07:22,440
the minus one to one range

203
00:07:22,440 --> 00:07:25,020
then we use Vector subtraction algorithm

204
00:07:25,020 --> 00:07:27,900
to denoise the signal

205
00:07:27,900 --> 00:07:29,460
and then after that we'll use an

206
00:07:29,460 --> 00:07:31,139
equalizer to amplify the response of

207
00:07:31,139 --> 00:07:33,840
weak frequencies

208
00:07:33,840 --> 00:07:35,520
now we will evaluate an attacking

209
00:07:35,520 --> 00:07:36,960
comparison to other attacks in this

210
00:07:36,960 --> 00:07:38,639
field

211
00:07:38,639 --> 00:07:41,099
the first attack we evaluated is is a

212
00:07:41,099 --> 00:07:43,560
experimentation sorry the first attack

213
00:07:43,560 --> 00:07:46,199
we evaluated in front of it is visual

214
00:07:46,199 --> 00:07:47,220
microphone

215
00:07:47,220 --> 00:07:49,740
and so we place the speakers five

216
00:07:49,740 --> 00:07:51,300
centimeters from the reflective object

217
00:07:51,300 --> 00:07:54,300
and the volume of the speakers was 95 DB

218
00:07:54,300 --> 00:07:55,919
which is very loud

219
00:07:55,919 --> 00:07:59,099
like in the visual microphone attack we

220
00:07:59,099 --> 00:08:01,139
then played an eftopping equipment 2.5

221
00:08:01,139 --> 00:08:03,720
meters from the object and we play the

222
00:08:03,720 --> 00:08:04,979
same sentences from the teammate

223
00:08:04,979 --> 00:08:07,139
repository that were used by visual

224
00:08:07,139 --> 00:08:08,940
microphone

225
00:08:08,940 --> 00:08:11,099
here we can see the spectrograms that we

226
00:08:11,099 --> 00:08:12,780
obtained from the original sentence and

227
00:08:12,780 --> 00:08:14,280
the optical measurements of values

228
00:08:14,280 --> 00:08:16,979
object that we used in this experiment

229
00:08:16,979 --> 00:08:18,240
you can obviously see that this

230
00:08:18,240 --> 00:08:20,039
spectacle of the original audio is at

231
00:08:20,039 --> 00:08:21,479
much higher quality

232
00:08:21,479 --> 00:08:23,580
but however I want to prove to you that

233
00:08:23,580 --> 00:08:25,259
the signals that we covered are still in

234
00:08:25,259 --> 00:08:27,300
good quality and you will be able to

235
00:08:27,300 --> 00:08:28,979
understand the content out of visual

236
00:08:28,979 --> 00:08:31,860
recovered sound

237
00:08:31,860 --> 00:08:34,620
so let's hear it I hope you will be able

238
00:08:34,620 --> 00:08:36,839
to tell me if you can't

239
00:08:36,839 --> 00:08:40,080
she had your dark suit greasy wash water

240
00:08:40,080 --> 00:08:42,620
all year

241
00:08:48,230 --> 00:08:51,320
[Music]

242
00:08:53,880 --> 00:08:57,140
do you feel anything

243
00:09:02,899 --> 00:09:06,420
okay so next we have added Attack under

244
00:09:06,420 --> 00:09:07,860
the same experiment that set up as long

245
00:09:07,860 --> 00:09:08,640
form

246
00:09:08,640 --> 00:09:10,680
I am just a reminder landform was

247
00:09:10,680 --> 00:09:13,260
presented in 20202 in use next by our

248
00:09:13,260 --> 00:09:16,140
group we demonstrated speech recovery by

249
00:09:16,140 --> 00:09:17,760
analyzing the vibrations of a hanging

250
00:09:17,760 --> 00:09:21,620
light bulb using a photodiode

251
00:09:22,620 --> 00:09:24,600
the experimental setup was duplicated as

252
00:09:24,600 --> 00:09:27,540
follows as follows we placed an eftop

253
00:09:27,540 --> 00:09:29,220
Dave stopping equipment at various

254
00:09:29,220 --> 00:09:30,240
distances

255
00:09:30,240 --> 00:09:32,640
first 15 meters then 25 meters and then

256
00:09:32,640 --> 00:09:34,140
35 meters

257
00:09:34,140 --> 00:09:36,000
the speakers were at a distance of 25

258
00:09:36,000 --> 00:09:38,399
centimeters on the reflective object and

259
00:09:38,399 --> 00:09:39,959
we'll play a statement made by phone

260
00:09:39,959 --> 00:09:41,519
president Donald Trump we are the

261
00:09:41,519 --> 00:09:42,839
speakers at the volume of a virtual

262
00:09:42,839 --> 00:09:46,940
meeting which is 75 decibels

263
00:09:47,420 --> 00:09:49,560
obtained out of the optical signal

264
00:09:49,560 --> 00:09:51,540
recovered from the Rubik's Cube from

265
00:09:51,540 --> 00:09:53,760
various distances you can see that the

266
00:09:53,760 --> 00:09:55,860
quality goes down as the distance goes

267
00:09:55,860 --> 00:09:58,880
up as expected

268
00:09:59,060 --> 00:10:03,018
hopefully you can hear them again

269
00:10:21,300 --> 00:10:23,700
so as you can hear the quality went down

270
00:10:23,700 --> 00:10:26,580
as long as the distance went up but it

271
00:10:26,580 --> 00:10:29,519
was still convenient to heal

272
00:10:29,519 --> 00:10:31,260
now let's review the takeaways from this

273
00:10:31,260 --> 00:10:32,519
talk

274
00:10:32,519 --> 00:10:34,680
so the primary takeaway of this talk is

275
00:10:34,680 --> 00:10:36,660
that as you can understand many shiny

276
00:10:36,660 --> 00:10:38,339
lightweight objects can serve as Optical

277
00:10:38,339 --> 00:10:40,680
implants and might be exploited by each

278
00:10:40,680 --> 00:10:42,360
robot will cover some

279
00:10:42,360 --> 00:10:44,220
most of the cases those those are

280
00:10:44,220 --> 00:10:46,200
completely innocent subjects and in some

281
00:10:46,200 --> 00:10:48,060
cases for example like Financial

282
00:10:48,060 --> 00:10:49,800
implants those objects should increase

283
00:10:49,800 --> 00:10:52,740
policy but can be exploited as the

284
00:10:52,740 --> 00:10:54,300
complete opposite as we showed in a

285
00:10:54,300 --> 00:10:56,399
research

286
00:10:56,399 --> 00:10:57,959
the second takeaway is that

287
00:10:57,959 --> 00:10:59,640
interestingly the area of speech is

288
00:10:59,640 --> 00:11:01,380
looping as advanced as advanced

289
00:11:01,380 --> 00:11:02,640
significantly in the last nine years

290
00:11:02,640 --> 00:11:05,040
ever since visual microphone shows how

291
00:11:05,040 --> 00:11:06,660
to use the open speech using high

292
00:11:06,660 --> 00:11:09,240
frequency camera and other attacks since

293
00:11:09,240 --> 00:11:11,820
then like lamp phone Lido phone the

294
00:11:11,820 --> 00:11:13,740
globe attack and today the little silver

295
00:11:13,740 --> 00:11:15,899
we expect that more Optical image

296
00:11:15,899 --> 00:11:17,579
dropping methods and the three line

297
00:11:17,579 --> 00:11:19,500
Optical sensors will be demonstrated in

298
00:11:19,500 --> 00:11:21,360
the next few years

299
00:11:21,360 --> 00:11:23,820
and for the last takeaway there are

300
00:11:23,820 --> 00:11:25,200
three interesting facts that I'd like to

301
00:11:25,200 --> 00:11:28,260
take into account so the first fact is

302
00:11:28,260 --> 00:11:30,660
that photo diodes are integrated into

303
00:11:30,660 --> 00:11:33,959
most smartphones nowadays and second is

304
00:11:33,959 --> 00:11:35,940
that smartphone manufacturers continue

305
00:11:35,940 --> 00:11:38,339
to increase the smartphone sampling rate

306
00:11:38,339 --> 00:11:40,200
the third fact is that obtaining

307
00:11:40,200 --> 00:11:42,060
obtaining the data from those and so

308
00:11:42,060 --> 00:11:44,040
does not require users permission

309
00:11:44,040 --> 00:11:45,180
currently

310
00:11:45,180 --> 00:11:47,279
all of those facts alone aren't very

311
00:11:47,279 --> 00:11:49,860
important or dangerous but combining

312
00:11:49,860 --> 00:11:51,839
them together we might see any an

313
00:11:51,839 --> 00:11:54,120
implementation of a new attack or

314
00:11:54,120 --> 00:11:56,940
malware or compromised applications for

315
00:11:56,940 --> 00:11:58,860
Speed Shift stopping via smartphone's

316
00:11:58,860 --> 00:12:01,440
lights and so in the near future

317
00:12:01,440 --> 00:12:03,420
and thank you very much for attending

318
00:12:03,420 --> 00:12:05,339
attending the talk I was I really

319
00:12:05,339 --> 00:12:07,079
appreciate it and now if you have any

320
00:12:07,079 --> 00:12:10,579
questions I will again sign them

321
00:12:16,320 --> 00:12:20,300
any questions to ask yes

322
00:12:22,019 --> 00:12:25,260
so I saw that you try different surfaces

323
00:12:25,260 --> 00:12:30,180
did you manage to understand the changes

324
00:12:30,180 --> 00:12:32,700
that you have with curved surfaces that

325
00:12:32,700 --> 00:12:36,060
compare to straight like flat surfaces

326
00:12:36,060 --> 00:12:39,000
when you are reflecting the signal

327
00:12:39,000 --> 00:12:42,839
and so yes light reflected from

328
00:12:42,839 --> 00:12:45,660
um flat surfaces was more

329
00:12:45,660 --> 00:12:48,120
centered into the photodiode like the

330
00:12:48,120 --> 00:12:50,339
photo that could receive the signal

331
00:12:50,339 --> 00:12:51,600
better from

332
00:12:51,600 --> 00:12:53,700
um flat surfaces because it was more

333
00:12:53,700 --> 00:12:56,880
centered and less um distributed

334
00:12:56,880 --> 00:12:59,220
and as you can see in the research it

335
00:12:59,220 --> 00:13:02,760
was it was much much easier to recover

336
00:13:02,760 --> 00:13:04,380
sound from the Rubik's Cube for example

337
00:13:04,380 --> 00:13:07,920
because it was very flat then from other

338
00:13:07,920 --> 00:13:10,380
objects and that's why also the nation

339
00:13:10,380 --> 00:13:12,000
Venetian blinds are very effective

340
00:13:12,000 --> 00:13:15,560
because it's very flat

341
00:13:20,120 --> 00:13:23,040
hello thanks for the talk I really enjoy

342
00:13:23,040 --> 00:13:26,100
your uh final Insight of uh the

343
00:13:26,100 --> 00:13:27,959
probability of turning the light sensor

344
00:13:27,959 --> 00:13:30,180
on smartphones into like during this

345
00:13:30,180 --> 00:13:32,880
attack although I have one question that

346
00:13:32,880 --> 00:13:35,040
is um like the camera like the common

347
00:13:35,040 --> 00:13:38,399
camera stuff has like the uh pinhole

348
00:13:38,399 --> 00:13:40,680
sort of thing but the light sensor

349
00:13:40,680 --> 00:13:43,139
itself does not have focus and it's like

350
00:13:43,139 --> 00:13:47,100
getting light from all the directions

351
00:13:47,100 --> 00:13:48,779
um so that means your significant noise

352
00:13:48,779 --> 00:13:50,760
ratio is going to be super low even if

353
00:13:50,760 --> 00:13:52,980
you have a very high sampling rate so uh

354
00:13:52,980 --> 00:13:55,320
during Invasion any sort of techniques

355
00:13:55,320 --> 00:13:56,820
that could kind of address that problem

356
00:13:56,820 --> 00:13:59,279
thank you

357
00:13:59,279 --> 00:14:01,260
um that's a good question

358
00:14:01,260 --> 00:14:04,560
and currently as you said the

359
00:14:04,560 --> 00:14:06,959
the sensor is not too good

360
00:14:06,959 --> 00:14:09,120
in the future when it will be good I'm

361
00:14:09,120 --> 00:14:12,180
not sure if if that I could still be

362
00:14:12,180 --> 00:14:14,100
able if you still be able to take this

363
00:14:14,100 --> 00:14:15,480
in this direction

364
00:14:15,480 --> 00:14:17,820
because of the distributed

365
00:14:17,820 --> 00:14:19,500
um area

366
00:14:19,500 --> 00:14:20,519
um

367
00:14:20,519 --> 00:14:23,700
I think the if the sensor does Focus for

368
00:14:23,700 --> 00:14:26,040
example on the light bulb always on the

369
00:14:26,040 --> 00:14:28,200
desk and focused on the light reflected

370
00:14:28,200 --> 00:14:29,160
from

371
00:14:29,160 --> 00:14:31,200
you don't need to focus on the surface

372
00:14:31,200 --> 00:14:33,000
of job deck of the object if you if

373
00:14:33,000 --> 00:14:34,740
there's that sort of just

374
00:14:34,740 --> 00:14:36,899
and receive the light reflected from the

375
00:14:36,899 --> 00:14:39,420
objects on the desk it can still work

376
00:14:39,420 --> 00:14:41,699
but it's an interesting question and I

377
00:14:41,699 --> 00:14:43,380
guess that we will be able to know it

378
00:14:43,380 --> 00:14:46,079
only when we have the rights and so then

379
00:14:46,079 --> 00:14:48,560
smartphones

380
00:14:49,380 --> 00:14:51,360
okay

381
00:14:51,360 --> 00:14:55,019
how low can the amplitude of the sound

382
00:14:55,019 --> 00:14:58,699
be before this uh doesn't work

383
00:14:58,699 --> 00:15:01,860
how long

384
00:15:01,860 --> 00:15:06,600
also we this is 95 DB right that's

385
00:15:06,600 --> 00:15:07,680
pretty loud

386
00:15:07,680 --> 00:15:09,779
that's very low that's the sound of

387
00:15:09,779 --> 00:15:12,480
Screams and that's what we used for

388
00:15:12,480 --> 00:15:14,699
comparison with the visual microphone

389
00:15:14,699 --> 00:15:16,740
that's the experiment that has updated

390
00:15:16,740 --> 00:15:19,440
so we've replicated it but we also put

391
00:15:19,440 --> 00:15:20,760
it in

392
00:15:20,760 --> 00:15:23,339
in the in the sound of a virtual meeting

393
00:15:23,339 --> 00:15:26,579
of 75 degrees in the experiment in

394
00:15:26,579 --> 00:15:27,959
comparison to

395
00:15:27,959 --> 00:15:29,160
um landform

396
00:15:29,160 --> 00:15:32,519
so 75 DBS is very similar to Virtual

397
00:15:32,519 --> 00:15:34,920
conversation maybe a little higher but

398
00:15:34,920 --> 00:15:37,199
we showed it from 75 degrees Yeah I

399
00:15:37,199 --> 00:15:38,459
believe that

400
00:15:38,459 --> 00:15:41,880
with maybe with better better setup and

401
00:15:41,880 --> 00:15:43,560
the equipment you might even do it from

402
00:15:43,560 --> 00:15:46,500
less than that but 70 75 degrees is

403
00:15:46,500 --> 00:15:49,760
there is what we proved

404
00:15:51,560 --> 00:15:54,000
oh and you do we have any further

405
00:15:54,000 --> 00:15:56,180
questions