12
Term Project Close-Set, Text-Dependent, Twelve(12) people, Speaker Identification Using ANN (Artificial Neural Networks) JAY DESAI KUANG-TAO CHIAO

Term Project Close-Set, Text-Dependent, Twelve(12) people, Speaker Identification Using ANN (Artificial Neural Networks) JAY DESAI KUANG-TAO CHIAO

Embed Size (px)

Citation preview

Page 1: Term Project Close-Set, Text-Dependent, Twelve(12) people, Speaker Identification Using ANN (Artificial Neural Networks) JAY DESAI KUANG-TAO CHIAO

Term Project

Close-Set, Text-Dependent, Twelve(12) people, Speaker Identification Using ANN

(Artificial Neural Networks)

JAY DESAI KUANG-TAO CHIAO

Page 2: Term Project Close-Set, Text-Dependent, Twelve(12) people, Speaker Identification Using ANN (Artificial Neural Networks) JAY DESAI KUANG-TAO CHIAO

Introduction

OverviewClosed Set/Open SetText Dependent/Text IndependentSpeaker Identification/Speaker

Verification

Page 3: Term Project Close-Set, Text-Dependent, Twelve(12) people, Speaker Identification Using ANN (Artificial Neural Networks) JAY DESAI KUANG-TAO CHIAO

System Architecture

Block Diagram

Page 4: Term Project Close-Set, Text-Dependent, Twelve(12) people, Speaker Identification Using ANN (Artificial Neural Networks) JAY DESAI KUANG-TAO CHIAO

Some Plots we obtained

Page 5: Term Project Close-Set, Text-Dependent, Twelve(12) people, Speaker Identification Using ANN (Artificial Neural Networks) JAY DESAI KUANG-TAO CHIAO

Short-time Energy

The 4 vowelsShort time energyLog plot

Page 6: Term Project Close-Set, Text-Dependent, Twelve(12) people, Speaker Identification Using ANN (Artificial Neural Networks) JAY DESAI KUANG-TAO CHIAO

Frame Extraction

3 frames/vowel168 Cepstral Coeff.

Page 7: Term Project Close-Set, Text-Dependent, Twelve(12) people, Speaker Identification Using ANN (Artificial Neural Networks) JAY DESAI KUANG-TAO CHIAO

Password

/u/ /i/ /æ/ /a/Why the choice of password?Vowel PlaneThe Phoneticians vowel trapezium

Page 8: Term Project Close-Set, Text-Dependent, Twelve(12) people, Speaker Identification Using ANN (Artificial Neural Networks) JAY DESAI KUANG-TAO CHIAO

Linear Predictive Coding

Why LP analysis?Feature ExtractionComputational aspectsLPC Cepstrum

Page 9: Term Project Close-Set, Text-Dependent, Twelve(12) people, Speaker Identification Using ANN (Artificial Neural Networks) JAY DESAI KUANG-TAO CHIAO

Artificial Neural NetworksArtificial Neural Networks

wki

θk

yk

nk

Page 10: Term Project Close-Set, Text-Dependent, Twelve(12) people, Speaker Identification Using ANN (Artificial Neural Networks) JAY DESAI KUANG-TAO CHIAO

Back Propagation

Page 11: Term Project Close-Set, Text-Dependent, Twelve(12) people, Speaker Identification Using ANN (Artificial Neural Networks) JAY DESAI KUANG-TAO CHIAO

Potential Applications

Meetings, Conferences, Conversations

Law enforcementSecurity applicationHuman-Machine InterfaceGender recognitionOthers

Page 12: Term Project Close-Set, Text-Dependent, Twelve(12) people, Speaker Identification Using ANN (Artificial Neural Networks) JAY DESAI KUANG-TAO CHIAO

Scope of Improvement

RobustnessAdditive NoiseCo-channel InterferenceIncreasing the number of users