Technical Seminar presentation on
Speech Recognition using DWT
Prepared By :
Why Speech Recognition ?????
Decrease the human intervention in different
processes or in other words automate them.
Security and privacy of documents.
DSP in Speech Recognition
DSP has its own functions in Mat lab for
Provides best quality of voice processing.
➌ Offers various options like
Challenges in speech Recognition
Human speech parameterized by different variables which vary
from speaker to speaker.
Speech signal consists of both vowels and consonants.
Speech signal is not stationary.
Languages vary the speech.
Frequency domain analysis.
Projects signals onto complex sines and cosines,
infinitely long signals
Carries both temporal location - like an impulse - and
frequency content - like a sinusoid.
Conception of wavelets
➊ Wavelets are localized waves and have their energy concentrated in
➋ “Wave” means Oscillatory and “let” means Quick decaying.
➌Difference between wave and wavelet :-
Conception of wavelets (contd.)
➍ Different types of wavelets
Wavelet families (a) Haar (b) Daubechies4 (c) Coiflet1 (d) Symlet2 (e) Meyer
(f) Morlet (g) Mexican Hat.
➊ Wavelet transform decomposes a signal into a set of basis
➋Wavelets are obtained from a single prototype wavelet y(t) called
mother wavelet by dilations and shifting.
1 t b
a ,b (t ) ( )
where a is the scaling parameter and b is the shifting parameter.
Wavelet Transform (contd.)
➌ Then what is DWT
Discrete wavelet transform (DWT), which transforms a discrete
time signal to a discrete wavelet representation.
Equation of a discrete mother wavelet
Wavelet Transform (contd.)
➊ DWT using Filter Bank theory
Daubechies wavelet transform :-
➊ An orthonormal, compactly supported family of wavelets.
➋ Calculated using the scaling functions and wavelet functions.
➌ Is the default wavelet transform present in mat lab.
32 point DWT
Speech Recognition :-
➊ Using Daubechies WT the signal is divided in to no of octaves.
➋ Analyze the different octaves and the characteristic octave is found
➌ Different characteristic properties are
magnitude of the amplitude
No of samples above a threshold level.
➍ Templates are created for different spoken words.
➎ Each time the input signal is compared with the templates.
Now Speech Recognition (contd):-
➏ An example of recognition of digits
Third octave of “two” and “three”
Speech Recognition (contd):-
➐ The Flow chart
➊ Wavelets proves to be an effective method in analyzing speech signals
that contain both steady state characteristics (vowels) and transient
➋ Better result can be got by finding even more differences in different
octaves between each of the digits and adding formants and pitch
determinations to the wavelet analysis.
➌ Future idea
Use of continuous wavelet transform which steps though the frequencies
and times continuously.