15
Seneff’s Auditory Seneff’s Auditory Model Model Miriam Cordero Ruiz Miriam Cordero Ruiz (SONY Advanced Technology Center (SONY Advanced Technology Center Stuttgart) Stuttgart) Leuven, july 2002 Leuven, july 2002

Seneff’s Auditory Model Miriam Cordero Ruiz (SONY Advanced Technology Center Stuttgart) Leuven, july 2002

  • View
    218

  • Download
    0

Embed Size (px)

Citation preview

Seneff’s Auditory ModelSeneff’s Auditory Model

Miriam Cordero RuizMiriam Cordero Ruiz(SONY Advanced Technology Center Stuttgart)(SONY Advanced Technology Center Stuttgart)

Leuven, july 2002Leuven, july 2002

Which is the best speech Which is the best speech recognizer?recognizer?

IntroductionIntroduction

•Auditory System

•Seneff’s Model

•Stage I

•Stage II

•Conclusions

Human Auditory SystemHuman Auditory System

Human Auditory SystemHuman Auditory System

Human Auditory SystemHuman Auditory Systemband fc(Hz) BW(Hz)

1 50 802 150 1003 250 1004 350 1005 450 1106 570 1207 700 1408 840 1509 1000 16010 1170 19011 1370 21012 1600 24013 1850 28014 2150 32015 2500 38016 2900 45017 3400 55018 4000 70019 4800 90020 5800 110021 7000 130022 8500 180023 10500 250024 13500 3500

Human Auditory SystemHuman Auditory System

t

t

Inner Hair CellsInner Hair Cells

Structure of the modelStructure of the model

CRITICAL BAND FILTER

BANK

HAIR CELL SYNAPSE

MODEL

ENVELOPE DETECTOR

SYNCHRONY DETECTOR

Mean rate spectrum

synchrony spectrum

STAGE I STAGE II STAGE III

Stage I: Stage I: Auditory Filter BankAuditory Filter Bank

40 channels (20 - 6700 Hz)

BW1channel=0,5 Barks

Design of the Auditory Filter Design of the Auditory Filter BankBank

INITIAL COMPLEX ZEROES

ZERO OF CASCADE

ZERO OF CASCADE

ZERO OF CASCADE

RESONATOR RESONATOR RESONATOR

CHANNEL 1 CHANNEL 2 CHANNEL 40

…….

f(Hz)

Stage IIStage IIPhysiological DataPhysiological Data ModelModel

< 1kHz< 1kHz

Stages I+IIStages I+II

CRITICAL BANDFILTER BANK

HALFWAVE RECTIFICATION

SHORT-TERM ADAPTATION

LOW PASS FILTER

RAPID AGC

STA

GE I

IS

TA

GE I

ResultsResults

Other Peripheral ModelsOther Peripheral Models

•Patterson-Meddis

Gammatone Filterbank

•Lyon’s Cochlear Model

Gammatone Filterbank

Adaptation Stage

ConclusionsConclusions

•Based on biological data

•Front-End for Speech Processing

Speech Recognition, Speaker ID, Localization….

•Better performance