A NEW FEATURE EXTRACTION MOTIVATED BY HUMAN EAR

A NEW FEATURE EXTRACTION MOTIVATED

BY HUMAN EAR

Amin Fazel

Sharif University of TechnologyHossein Sameti, S. K. Ghiathi

February 2005

Department of Computer Engineering 2/26Thursday, February 03, 2005

Introduction

Physiological basis in the human auditory system

Modeling of the basilar membrane and hair cells

Experimental results

Summary and conclusions

Outline


Introduction Speech is #1 real-time communication

medium among humans.

Advantages of voice interface to machines: Hands-free operation Speed Ease of use


Introduction Human is a

high-performance existence proof for speech recognition in noisy environments.

Wall Street Journal/Broadcast news readings, 5000 words

Untrained human listeners vs. Cambridge HTK LVCSR system


Physiological Basis


Physiological Basis

The semicircular canals are the body's balance organs.

Hair cells, in the canals, detect movements of the fluid in the canals caused by angular acceleration

The canals are connected to the auditory nerve.

Semicircular Canals

Cochlea

Inner Ear


Physiological Basis

The inner ear structure called the cochlea is a snail-shell like structure divided into three fluid-filled parts.

Two are canals (Scala tympani and Scala Vestibuli) for the transmission of pressure and in the third is the sensitive organ of Corti, which detects pressure impulses and responds with electrical impulses which travel along the auditory nerve to the brain

Semicircular Canals

Cochlea

Inner Ear


Physiological Basis

The organ of Corti can be thought of as the body's microphone. Perception of pitch and perception of loudness is connected with this

organ.

It is situated on the basilar membrane in the cochlea duct It contains inner hair cells and outer hair cells. There are some 16,000 -20,000 of the hair cells distributed along the

basilar membrane.

Vibrations of the oval window causes the cochlear fluid to vibrate. This causes the Basilar membrane to vibrate thus producing a traveling

wave. This causes the bending of the hair cells which produces generator

potentials If large enough will stimulate the fibers of the auditory nerve to produce

action potentials The outer hair cells amplify vibrations of the basilar membrane

Semicircular Canals

Cochlea

Inner Ear


Modeling of BM and Hair Cells

Different parts of basilar membrane and hair cells are sensitive to different frequencies of input signal.


Modeling of BM and Hair Cells Since corporation of basilar membrane and hair

cells changes all frequencies of speech into mechanical energy, with good approximation, we can discretely represent basilar membrane and hair cells as forced damped oscillators with different natural frequencies.


Modeling of BM and Hair Cells We stimulate these oscillators with input sound

In this simulation we have an oscillating particle which is always pulled by a force towards the center of oscillation

Displacement of the article from the center of oscillation is shown by x and the inward force is equal to –kx.

k is the constant for each oscillator

20mk

constant


Modeling of BM and Hair Cells Since we have a foreign force (posed by sound),

we can no further use those standard equations which assume the energy of system is constant. If we don't consider the effect of friction, the energy of system will not decrease and it becomes instable. So we must add a force in opposite direction of movement. Since the direction of movement is determined by v (velocity), the friction force is –bv

Viewing each diapason as a filter

Q

mb 0

Bandwidth


Modeling of BM and Hair Cells We model the state of each oscillator with

the pair [x v], where x is the displacement and v is the velocity of particle

Where ∆t is the inverse of sampling frequency

a

v

x

t

t

v

xold

old

new

new

10

01


Modeling of BM and Hair Cells The particle is imposed by three forces:

The diapason itself pulls the particle by force –kx

The sound imposes a foreign force, say Fexternal

To compute Fexternal from the current sample we use the value of sample itself as the external force

The friction opposes to the movement by force –bv


Modeling of BM and Hair Cells Now we can compute a, using the

following formula

For using this model in feature extraction After calculation of the energy for each of

these oscillators, we use them as feature vectors in ASR systems

m

kxbvFa prpr

22

2

1

2

1kxmvE


Experimental results We transform a speech with our human based

model and compare it to spectrum domain of this speech

These two transformations have little differences


Experimental results This comparing shows that this human

based model can be used impressively in ASR systems.

In addition, this method can be used as an effective and quick signal transformation instead of FFT or wavelet in various tasks.


ASR Experiments The feature extraction algorithm proposed

for speech recognition were tested on a English digit database For training we use 1386 digit sequences

spoken by 18 speakers

In testing phase we use 200 digit sequences that uttered by speakers out of training database

The testing database split to four groups of 50 sequences and four types of noises added to these groups


ASR Experiments Recognition is performed using HTK

16 emitting states and three mixture continuous HMM model

3-state silence model Single state inter-digit pause model

In the reference experiments, MFCC_0_D_A is used Consists of 13 standard cepstral coefficients including

C0 augmented with first and second derivations of them

MFCC features were generated by applying a Hamming window of size 25 ms and overlap 10 ms to the same pre-emphasized 23-channel Mel-scale filterbank.

The cepstral features were obtained from DCT of log-energy over the 23 frequency channels.


ASR Experiments Car Noise

Comparing of MFCC and HEFE for Car Noise

0102030405060708090

100

20dB 15dB 10dB 5dB 0dB -5dBSNR. dB

Wo

rd e

rro

r R

ate

%

MFCC

HEFE


ASR Experiments Exhibition Noise

Comparing of MFCC and HEFE for Exhibition Noise

0

20

40

60

80

100

20dB 15dB 10dB 5dB 0dB -5dBSNR. dB

Wo

rd e

rro

r R

ate

%

MFCC

HEFE


ASR Experiments Babble Noise

Comparing of MFCC and HEFE for Babble Noise

0

20

40

60

80

100

120

20dB 15dB 10dB 5dB 0dB -5dB

SNR. dB

Wo

rd e

rro

r R

ate

%

MFCC

HEFE


ASR Experiments Subway Noise

Comparing of MFCC and HEFE for Subway Noise

0

20

40

60

80

100

20dB 15dB 10dB 5dB 0dB -5dB

SNR. dB

Wo

rd e

rro

r R

ate

%

MFCC

HEFE


ASR Experiments For all contaminated speech, HEFE shows

superior performance for all noise types at most SNR levels.

For babble noise, HEFE demonstrates significantly better performance than MFCC.

For subway noise, improvements by the HEFE are least significant, but still noticeable.


Summary In this paper we have introduced a simple

model for basilar membrane and hair calls based on physiological basis

We use this model for feature extraction in ASR systems

These features significantly outperform MFCC features at babble noise

Thank you!

Documents

A NEW FEATURE EXTRACTION MOTIVATED BY HUMAN EAR