DEVELOPMENT OF A NOVEL VOICE

VERIFICATION SYSTEM USING

WAVELETS
AIM OF THE PROJECT
The aim of the project is to study The Voice
Recognition Using Wavelet Feature
Extraction employ wavelets in voice
recognition for studying the dynamic
properties and characteristics of the voice
signal.
INTRODUCTION
The developed voice recognition system is word
independant voice verification system combining
the RASTA and LPC.
The voice signal is filtered using the special
purpose voice signal filter using the Relative
Spectral Algorithm (RASTA)
The signals are denoised and decomposed to
derive the wavelet coefficients and thereby a
statistical computation is carried out.
CONT....
the accuracy of the verifying sample individual
voice to his own voice is quite high (around 75%
to 80%).
The reliability of the signal verification is
strengthened by combining entailments from
these two completely different aspectsof the
individual voice.
CONT....
For voice comparison purposes four out five
individuals are verified and the results show
higher percentage of accuracy.
The accuracy of the system can be improved by
incorporating advanced pattern recognition
techniques such as Hidden Markov Model
(HMM).
WAVELETS
Data


Time domain



Frequency domain



FOURIER TRANSFORMS
The Great Father Fourier - Fourier Transforms



Any Periodic function can be expressed as
linear combination of basic trigonometric
functions
(Basis functions used are sine and cosine)


}
}

=
=
df e f X t x
dt e t x f X
ift
ift
t
t
2
2
) ( ) (
) ( ) (
COSINE AND SINE PARTS
Real part
and imaginary
part

) sin( ) cos( 2 / ) (
0
ft b ft a a t x
f f
+ + =

DRAWBACKS
Integration from -inf to +inf
Gives frequency content of total time series but temporal
information is lost!
SHORT TIME FOURIER TRANSFORMS
Same as usual Fourier transforms, but data is modified by
multiplication with a window function
Only part of data at a time is taken and processed
DRAWBACKS OF STFT
Frequency and time resolutions are fixed
(Wider the window width, lesser the time resolution
and more the frequency resolution and vice versa)

As frequency resolution increases, time
resolution decreases uncertainty principle
Desired: Good time resolution at high
frequencies and good frequency
resolution at low frequencies!

WHAT ARE WAVELETS?
A small wave
Extends to finite interval

SOME MATHEMATICAL EXPRESSIONS
}

|
.
|

\
|

= dt
s
t
t x
s
s W
t
t ) (
1
) , (
x(t) actual time series
(t) wavelet function
( ) <
=
}
}

dt t
dt t
2
) (
0 ) (

Integrable and limited to finite

region
Total energy finite
TYPICAL PICTURE
1D WAVELET TRANSFORM
RASTA (RELATIVE SPECTRAL ALGORITHM)
This method works by applying a band-pass filter
to the energy in each frequency sub-band in order
to smooth over short-term noise variations and to
remove any constant offset.
stationary noises are often detected. Stationary
noises are noises that are present for the
fullperiod of a certain signal and does not have
diminishing feature
CONT..
The assumption that needs to bemade is that the
noise varies slowly with respect to speech.
This makes the RASTA a perfect tool to be
included in the initial stages of voice signal
filtering to remove stationary noises
The stationary noises that are identified are noises
in the frequency range of 1Hz - 100Hz.
SYSTEM IMPLEMENTATION
RESULTS
CONT...
CONCLUSION
The Voice Recognition Using Wavelet Feature
Extraction employ wavelets in voice recognition for
studying the dynamic properties and characteristics
ofthe voice signal. The voice recognition system that is
developed is word dependant voice verification system
used to verify the identity of an individual based on
their own voice signal using the statistical computation,
formant estimation and wavelet energy.. By using the
fifty preloaded voice signals from five individuals, the
verification tests have been carried and an accuracy rate
of approximately 80 % has been achieved
THANK YOU