Speaker Recognition and Speaker Recognition and the ETSI Standard - - PowerPoint PPT Presentation

Speaker Recognition and Speaker Recognition and the ETSI Standard the ETSI Standard Distributed Speech Distributed Speech Recognition Front Recognition Front-

• End

End

Charles Broun William Campbell

Motorola Labs Human Interface Lab Tempe, Arizona, USA

David Pearce Holly Kelleher

Motorola Limited Basingstoke, UK

2

Outline Outline Outline

• Background
• Speaker Verification

– Embedded Process – Distributed Process

• Distributed Speech Recognition (DSR)
• Classifier
• Experimental Setup
• Results
• Conclusion

3

Background Background Background

Motivation

• Issues with Embedded Solutions

Mobile devices do not have the necessary memory or battery capacity Updating software requires access to each device Multiple devices may contain different speaker models

• Potential Benefits of Distributed Solutions

Server supports computation and memory requirements Software updates are handled in a single location A single speaker model may support multiple mobile devices – enabling a ‘portable’ interface

• Distributed Speech Recognition (DSR) Standard

This standard addresses the above issues for speech recognition Can work on this standard be leveraged for speaker verification?

4

Speaker Verification Speaker Verification Speaker Verification

Embedded Process

• Feature extractor and classifier typically combined

into a proprietary solution

• Can jointly optimize both components
• Target system must support computation & memory

requirements of both components

Feature Extractor Input Speech Data Score Speaker Model Compare to Threshold, T Accept >T Reject <T Classifier

5

Speaker Verification Speaker Verification Speaker Verification

Distributed Process

• Feature extractor is standardized
• Cannot jointly optimize both components
• Client only supports computational & memory

requirements of feature extractor

• Server supports higher load of classifier

Feature Extractor Input Speech Data Score Speaker Model Compare to Threshold, T Accept >T Reject <T Classifier Wireless Channel

6

DSR DSR DSR

Background of DSR Standard

• Motivation of Standard Front-End

Potential benefits of distributed solutions for speech recognition Eliminates voice/vocoder channel mismatch

• Activities

European Telecommunications Standards Institute (ETSI) Aurora Working Group within ETSI First standard published in February 2000

7

DSR DSR DSR

ETSI Standard DSR System Concept

• Terminal front-end targeted to mobile devices
• Features transmitted over a low-error data channel
• Speech recognizer runs on high power server

Server DSR Back-End Terminal DSR Front-End

Parameterisation M el-Cepstrum Compression Split V Q Frame Structure & Error Protection Error Detection & M itigation Decompression

Recognition W ireless Data Channel – 4.8 kbit/s

8

DSR DSR DSR

ETSI Standard DSR Front-End

• Feature set consists of 12 mel-cepstum coefficient, logE, C0
• Quantization supports a data rate of 4800 b/s
• Error protection supports robustness to transmission errors

ADC Offcom Framing PE W FFT MF LOG DCT logE Feature Compression Bit Stream Formatting Input Speech To Transmission Channel Abbreviations: ADC Offcom PE logE W FFT MF LOG DCT Analog-to-digital conversion Offset compensation Pre-emphasis Energy measure computation Windowing Fast Fourier transform Mel-filtering Non-linear transform Discrete cosine transform

9

Classifier Classifier Classifier

Polynomial Classifier

• Compute the polynomial basis vector
• Apply a polynomial discriminant

function

• Compute the score as the average

across all frames

) ( ) , ( x p w w x

t

d =

∑ ∑

= =

= =

M k k t M k k t

M M s

1 1

) ( 1 ) ( 1 x p w x p w

[ ]

[ ]

t t

x x x x x x K x x

2 2 2 1 2 1 2 1 2 1

1 ) ( 2 and Given = = = x p x

DSR Feature Vectors Score Speaker Model w Discriminant Function d(x,w) Average

Σ

k

x s

Polynomial Basis Vector p(x)

10

Experimental Setup Experimental Setup Experimental Setup

YOHO Database

• 138 speakers
• Enrollment

– 4 sessions – 24 phrases – “23-45-56”

• Testing

– 10 sessions – 4 phrases – “45-23-56”

Speaker Verification System

• Classifier: 3rd order polynomial
• Features: 12 MFCCs from the DSR front-end
• Channel: GSM bit-error masks

11

Results Results Results

Performance

Average Equal Error Rate (%) for a 1-Phrase Test

1.66 1.27 1.22 1.22

• EP2

EP3 EP2 EP1 Error

• Free

Un- quantized Verify Enroll EP3 EP1 Error-Free Unquantized 1.70 1.30 1.26 1.26

• 1.67

1.26 1.22 1.22

• 1.67

1.26 1.22 1.22

• 1.18

12

Conclusion Conclusion Conclusion

Demonstrated that the ETSI Standard Distributed Speech Recognition Front-End is viable for speaker verification