did i say that?? speech lab spring 2009 february 03, 09 1 montgomery college did i say that? did i...

17
d I say that?? Speech Lab Spring 2009 February 03, February 03, 09 09 1 Montgomery Montgomery College College Did I Say That? Did I Say That? Automatic Keyword Spotting Using Crosscorrelation Uchechukwu Abanulo In collaboration with Temple University Speech Lab Funded by the US Air Force Research Lab, Rome, NY Uche O. Abanulo Physics, Engineering And Geosciences

Upload: roberta-martin

Post on 17-Jan-2018

286 views

Category:

Documents


0 download

DESCRIPTION

Did I say that?? Speech Lab Spring 2009 February 03, 09 3 Montgomery College Research Goal Applications Method Preliminary Results

TRANSCRIPT

Page 1: Did I say that?? Speech Lab Spring 2009 February 03, 09 1 Montgomery College Did I Say That? Did I Say That? Automatic Keyword Spotting Using Crosscorrelation

Did I say that?? Speech Lab Spring 2009

February 03, 09February 03, 0911

Montgomery Montgomery CollegeCollege

Did I Say That?Did I Say That?Automatic Keyword Spotting Using

Crosscorrelation

Uchechukwu Abanulo

In collaboration with Temple University Speech LabFunded by the US Air Force Research Lab, Rome, NY

Uche O. AbanuloPhysics, Engineering

And Geosciences

Page 2: Did I say that?? Speech Lab Spring 2009 February 03, 09 1 Montgomery College Did I Say That? Did I Say That? Automatic Keyword Spotting Using Crosscorrelation

Did I say that?? Speech Lab Spring 2009

February 03, 09February 03, 0922

Montgomery Montgomery CollegeCollege

Presentation OutlinePresentation Outline

Uche O. AbanuloPhysics, Engineering

And Geosciences

• Research Goal• Applications of Research• Method• Preliminary Results

Page 3: Did I say that?? Speech Lab Spring 2009 February 03, 09 1 Montgomery College Did I Say That? Did I Say That? Automatic Keyword Spotting Using Crosscorrelation

Did I say that?? Speech Lab Spring 2009

February 03, 09February 03, 0933

Montgomery Montgomery CollegeCollege

Research GoalResearch Goal

Applications

Method

Preliminary Results

Page 4: Did I say that?? Speech Lab Spring 2009 February 03, 09 1 Montgomery College Did I Say That? Did I Say That? Automatic Keyword Spotting Using Crosscorrelation

Did I say that?? Speech Lab Spring 2009

February 03, 09February 03, 0944

Montgomery Montgomery CollegeCollege

Research Goal Research Goal

• Keyword or Key-phrase detection– Did the speaker say __________?

• Keyword or Key-phrase Identification– What portion of the utterance contains

___________?

Research Goal

Applications

Method

Preliminary Results

Page 5: Did I say that?? Speech Lab Spring 2009 February 03, 09 1 Montgomery College Did I Say That? Did I Say That? Automatic Keyword Spotting Using Crosscorrelation

Did I say that?? Speech Lab Spring 2009

February 03, 09February 03, 0955

Montgomery Montgomery CollegeCollege

System Specifications System Specifications

• Speaker independence• Minimum utterance length – 2 seconds• Noise/Interference Robustness• Confidence Level Outputs

Research Goal

Applications

Method

Preliminary Results

Page 6: Did I say that?? Speech Lab Spring 2009 February 03, 09 1 Montgomery College Did I Say That? Did I Say That? Automatic Keyword Spotting Using Crosscorrelation

Did I say that?? Speech Lab Spring 2009

February 03, 09February 03, 0966

Montgomery Montgomery CollegeCollege

IllustrationIllustration

Listening DeviceListening Device

Research Goal

Applications

Method

Preliminary Results

Page 7: Did I say that?? Speech Lab Spring 2009 February 03, 09 1 Montgomery College Did I Say That? Did I Say That? Automatic Keyword Spotting Using Crosscorrelation

Did I say that?? Speech Lab Spring 2009

February 03, 09February 03, 0977

Montgomery Montgomery CollegeCollege

Research Goal

Applications

Method

Preliminary Results

Applications

Page 8: Did I say that?? Speech Lab Spring 2009 February 03, 09 1 Montgomery College Did I Say That? Did I Say That? Automatic Keyword Spotting Using Crosscorrelation

Did I say that?? Speech Lab Spring 2009

February 03, 09February 03, 0988

Montgomery Montgomery CollegeCollege

Eliminate manual listening to terrorist or ‘enemy’ Eliminate manual listening to terrorist or ‘enemy’ conversationsconversations

Homeland SecurityHomeland Security

• Automatically detect when targeted persons or groups utter certain flag words

Research Goal

Applications

Method

Preliminary Results

Page 9: Did I say that?? Speech Lab Spring 2009 February 03, 09 1 Montgomery College Did I Say That? Did I Say That? Automatic Keyword Spotting Using Crosscorrelation

Did I say that?? Speech Lab Spring 2009

February 03, 09February 03, 0999

Montgomery Montgomery CollegeCollege

Commercial Commercial

• Automatically search through speeches of important personnel for certain words or phrases

• Automated response systems Research Goal

Applications

Method

Preliminary Results

Page 10: Did I say that?? Speech Lab Spring 2009 February 03, 09 1 Montgomery College Did I Say That? Did I Say That? Automatic Keyword Spotting Using Crosscorrelation

Did I say that?? Speech Lab Spring 2009

February 03, 09February 03, 091010

Montgomery Montgomery CollegeCollege

Criminal Activity DetectionCriminal Activity Detection• Monitoring inmate conversations

• Detection of use of unpermitted words• Automatic searches for flag words

Research Goal

Applications

Method

Preliminary Results

Page 11: Did I say that?? Speech Lab Spring 2009 February 03, 09 1 Montgomery College Did I Say That? Did I Say That? Automatic Keyword Spotting Using Crosscorrelation

Did I say that?? Speech Lab Spring 2009

February 03, 09February 03, 091111

Montgomery Montgomery CollegeCollege

MethodResearch Goal

Applications

Method

Preliminary Results

Page 12: Did I say that?? Speech Lab Spring 2009 February 03, 09 1 Montgomery College Did I Say That? Did I Say That? Automatic Keyword Spotting Using Crosscorrelation

Did I say that?? Speech Lab Spring 2009

February 03, 09February 03, 091212

Montgomery Montgomery CollegeCollege

Cross correlationCross correlation• Measure of similarity between two

signals• Two signals compared by

– Sliding one signal by a certain time lag– Multiplying both the overlapping regions– Repeating the process and adding the

products until there is no more overlap• If both signals are exactly the same,

there’s a maximum peak at the time = 0, and the rest of the correlation signals tapers of to zero

Research Goal

Applications

Method

Preliminary Results

Page 13: Did I say that?? Speech Lab Spring 2009 February 03, 09 1 Montgomery College Did I Say That? Did I Say That? Automatic Keyword Spotting Using Crosscorrelation

Did I say that?? Speech Lab Spring 2009

February 03, 09February 03, 091313

Montgomery Montgomery CollegeCollege

Cross correlationCross correlation

Research Goal

Applications

Method

Preliminary Results

0 100 200 300 400 500 600 700 800 900 1000-5

0

5

Y

Illustration of Correlation

0 100 200 300 400 500 600 700 800 900 10000

0.5

1X

-1000 -800 -600 -400 -200 0 200 400 600 800 1000-1000

0

1000

XC

orr(y

)

-1000 -800 -600 -400 -200 0 200 400 600 800 10000

200

400

XC

orr(x

)

-1000 -800 -600 -400 -200 0 200 400 600 800 1000-50

0

50

Lag

XC

orr(x

,y)

Page 14: Did I say that?? Speech Lab Spring 2009 February 03, 09 1 Montgomery College Did I Say That? Did I Say That? Automatic Keyword Spotting Using Crosscorrelation

Did I say that?? Speech Lab Spring 2009

February 03, 09February 03, 091414

Montgomery Montgomery CollegeCollege

ApproachApproach

Research Goal

Applications

Method

Preliminary Results

0 1 2 3 4 5 6

x 104

-0.5

-0.4

-0.3

-0.2

-0.1

0

0.1

0.2

0.3

0.4

0.5

0 0.5 1 1.5 2 2.5 3 3.5 4

x 104

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1. Let the length of the keyword or phrase be n. The cross correlation of the keyword and the first n samples of the utterance is computed.

xcorrxcorr

0 1 2 3 4

x 104

-20

0

20

Max power is not Max power is not around zero lag – not around zero lag – not position of keywordposition of keyword

2. Observe position of peak to see if it’s around the zero lag.

Yes: KeywordNo: Not keyword

3. Shift observed portion by a small amount and repeat process

If a portion is reached where If a portion is reached where the peak is close to the zero the peak is close to the zero

lag, then that’s where the lag, then that’s where the keyword is. keyword is.

If not, the utterance does not If not, the utterance does not contain the keyword.contain the keyword.

Page 15: Did I say that?? Speech Lab Spring 2009 February 03, 09 1 Montgomery College Did I Say That? Did I Say That? Automatic Keyword Spotting Using Crosscorrelation

Did I say that?? Speech Lab Spring 2009

February 03, 09February 03, 091515

Montgomery Montgomery CollegeCollege

DemoResearch Goal

Applications

Method

Preliminary Results

Page 16: Did I say that?? Speech Lab Spring 2009 February 03, 09 1 Montgomery College Did I Say That? Did I Say That? Automatic Keyword Spotting Using Crosscorrelation

Did I say that?? Speech Lab Spring 2009

February 03, 09February 03, 091616

Montgomery Montgomery CollegeCollege

ResultsResults

Research Goal

Applications

Method

Preliminary Results

0

10

20

30

40

50

60

70

80

Hits False Alarms

Page 17: Did I say that?? Speech Lab Spring 2009 February 03, 09 1 Montgomery College Did I Say That? Did I Say That? Automatic Keyword Spotting Using Crosscorrelation

Did I say that?? Speech Lab Spring 2009

February 03, 09February 03, 091717

Montgomery Montgomery CollegeCollege

Uche O. AbanuloPhysics, Engineering

And Geosciences