Speech recognition-using-wavelet-transform

MAIN PROJECT ‘10 SPEECH RECOGNITION USING WAVELET TRANSFORM

1

1. INTRODUCTION

Automatic speech recognition (ASR) aims at converting spoken language to text.

Scientists all over the globe have been working under the domain, speech recognition for last

many decades. This is one of the intensive areas of research. Recent advances in soft

computing techniques give more importance to automatic speech recognition. Large variation

in speech signals and other criteria like native accent and varying pronunciations makes the

task very difficult. ASR is hence a complex task and it requires more intelligence to achieve a

good recognition result.

Speech recognition is currently used in many real-time applications, such as cellular

telephones, computers, and security systems. However, these systems are far from perfect in

correctly classifying human speech into words. Speech recognizers consist of a feature

extraction stage and a classification stage. The parameters from the feature extraction stage

are compared in some form to parameters extracted from signals stored in a database or

template. The parameters could be fed to a neural network.

Speech word recognition systems commonly carry out some kind of classification

recognition based on speech features which are usually obtained via Fourier Transforms

(FTs), Short Time Fourier Transforms (STFTs), or Linear Predictive Coding techniques.

However, these methods have some disadvantages. These methods accept signal stationarity

within a given time frame and may therefore lack the ability to analyze localized events

correctly. The wavelet transform copes with some of these problems. Other factors

influencing the selection of Wavelet Transforms (WT) over conventional methods include

their ability to determine localized features. Discrete Wavelet Transform method is used for

speech processing.

www.final-yearprojects.co.cc | www.troubleshoot4free.com/fyp/


2

2. LITERATURE SURVEY

Designing a machine that mimics human behavior, particularly the capability of

speaking naturally and responding properly to spoken language, has intrigued engineers and

scientists for centuries. Since the 1930s, when Homer Dudley of Bell Laboratories proposed a

system model for speech analysis and synthesis, the problem of automatic speech recognition

has been approached progressively, from a simple machine that responds to a small set of

sounds to a sophisticated system that responds to fluently spoken natural language and takes

into account the varying statistics of the language in which the speech is produced. Based on

major advances in statistical modeling of speech in the 1980s, automatic speech recognition

systems today find widespread application in tasks that require a human-machine interface,

such as automatic call processing in the telephone network and query-based information

systems that do things like provide updated travel information, stock price quotations,

weather reports, etc.

Speech is the primary means of communication between people. For reasons

ranging from technological curiosity about the mechanisms for mechanical realization of

human speech capabilities, to the desire to automate simple tasks inherently requiring human-

machine interactions, research in automatic speech recognition (and speech synthesis) by

machine has attracted a great deal of attention over the past five decades.

The desire for automation of simple tasks is not a modern phenomenon, but one

that goes back more than one hundred years in history. By way of example, in 1881

Alexander Graham Bell, his cousin Chichester Bell and Charles Sumner Tainter invented a

recording device that used a rotating cylinder with a wax coating on which up-and-down

grooves could be cut by a stylus, which responded to incoming sound pressure (in much the

same way as a microphone that Bell invented earlier for use with the telephone). Based on

this invention, Bell and Tainter formed the Volta Graphophone Co. in 1888 in order to

manufacture machines for the recording and reproduction of sound in office environments.

The American Graphophone Co., which later became the Columbia Graphophone Co.,

acquired the patent in 1907 and trademarked the term “Dictaphone.” Just about the same



3

time, Thomas Edison invented the phonograph using a tinfoil based cylinder, which was

subsequently adapted to wax, and developed the “Ediphone” to compete directly with

Columbia. The purpose of these products was to record dictation of notes and letters for a

secretary (likely in a large pool that offered the service) who would later type them out

(offline), thereby circumventing the need for costly stenographers.

This turn-of-the-century concept of “office mechanization” spawned a range of

electric and electronic implements and improvements, including the electric typewriter,

which changed the face of office automation in the mid-part of the twentieth century. It does

not take much imagination to envision the obvious interest in creating an “automatic

typewriter” that could directly respond to and transcribe a human‟s voice without having to

deal with the annoyance of recording and handling the speech on wax cylinders or other

recording media.

A similar kind of automation took place a century later in the 1990‟s in the area

of “call centers.” A call center is a concentration of agents or associates that handle telephone

calls from customers requesting assistance. Among the tasks of such call centers are routing

the in-coming calls to the proper department, where specific help is provided or where

transactions are carried out. One example of such a service was the AT&T Operator line

which helped a caller place calls, arrange payment methods, and conduct credit card

transactions. The number of agent positions (or stations) in a large call center could reach

several thousand Automatic speech recognition.

From Speech Production Models to Spectral Representations

Attempts to develop machines to mimic a human‟s speech communication

capability appear to have started in the 2nd

half of the 18th

century. The early interest was not

on recognizing and understanding speech but instead on creating a speaking machine,

perhaps due to the readily available knowledge of acoustic resonance tubes which were used

to approximate the human vocal tract. In 1773, the Russian scientist Christian Kratzenstein, a

professor of physiology in Copenhagen, succeeded in producing vowel sounds using

resonance tubes connected to organ pipes. Later, Wolfgang von Kempelen in Vienna

constructed an “Acoustic-Mechanical Speech Machine” (1791) and in the mid-1800's Charles

Wheatstone [6] built a version of von Kempelen's speaking machine using resonators made of



4

leather, the configuration of which could be altered or controlled with a hand to produce

different speech-like sounds.

During the first half of the 20th

century, work by Fletcher [8] and others at Bell

Laboratories documented the relationship between a given speech spectrum (which is the

distribution of power of a speech sound across frequency), and its sound characteristics as

well as its intelligibility, as perceived by a human listener. In the 1930‟s Homer Dudley,

influenced greatly by Fletcher‟s research, developed a speech synthesizer called the VODER

(Voice Operating Demonstrator), which was an electrical equivalent (with mechanical

control) of Wheatstone‟s mechanical speaking machine. Dudley‟s VODER which consisted

of a wrist bar for selecting either a relaxation oscillator output or noise as the driving signal,

and a foot pedal to control the oscillator frequency (the pitch of the synthesized voice). The

driving signal was passed through ten band pass filters whose output levels were controlled

by the operator‟s fingers. These ten band pass filters were used to alter the power distribution

of the source signal across a frequency range, thereby determining the characteristics of the

speech-like sound at the loudspeaker. Thus to synthesize a sentence, the VODER operator

had to learn how to control and “play” the VODER so that the appropriate sounds of the

sentence were produced. The VODER was demonstrated at the World Fair in New York City

in 1939 and was considered an important milestone in the evolution of speaking machines.

Speech pioneers like Harvery Fletcher and Homer Dudley firmly established the

importance of the signal spectrum for reliable identification of the phonetic nature of a speech

sound. Following the convention established by these two outstanding scientists, most

modern systems and algorithms for speech recognition are based on the concept of

measurement of the (time-varying) speech power spectrum (or its variants such as the

cepstrum), in part due to the fact that measurement of the power spectrum from a signal is

relatively easy to accomplish with modern digital signal processing techniques.

Early Automatic Speech Recognizers

Early attempts to design systems for automatic speech recognition were mostly

guided by the theory of acoustic-phonetics, which describes the phonetic elements of speech

(the basic sounds of the language) and tries to explain how they are acoustically realized in a

spoken utterance. These elements include the phonemes and the corresponding place and

manner of articulation used to produce the sound in various phonetic contexts. For example,



5

in order to produce a steady vowel sound, the vocal cords need to vibrate (to excite the vocal

tract), and the air that propagates through the vocal tract results in sound with natural modes

of resonance similar to what occurs in an acoustic tube. These natural modes of resonance,

called the formants or formant frequencies, are manifested as major regions of energy

concentration in the speech power spectrum. In 1952, Davis, Biddulph, and Balashek of Bell

Laboratories built a system for isolated digit recognition for a single speaker, using the

formant frequencies measured (or estimated) during vowel regions of each digit. These

trajectories served as the “reference pattern” for determining the identity of an unknown digit

utterance as the best matching digit.

In other early recognition systems of the 1950‟s, Olson and Belar of RCA

Laboratories built a system to recognize 10 syllables of a single talker and at MIT Lincoln

Lab, Forgie and Forgie built a speaker-independent 10-vowel recognizer. In the 1960‟s,

several Japanese laboratories demonstrated their capability of building special purpose

hardware to perform a speech recognition task. Most notable were the vowel recognizer of

Suzuki and Nakata at the Radio Research Lab in Tokyo, the phoneme recognizer of Sakai and

Doshita at Kyoto University, and the digit recognizer of NEC Laboratories [14]. The work of

Sakai and Doshita involved the first use of a speech segmenter for analysis and recognition of

speech in different portions of the input utterance. In contrast, an isolated digit recognizer

implicitly assumed that the unknown utterance contained a complete digit (and no other

speech sounds or words) and thus did not need an explicit “segmenter.” Kyoto University‟s

work could be considered a precursor to a continuous speech recognition system.

In another early recognition system Fry and Denes, at University College in

England, built a phoneme recognizer to recognize 4 vowels and 9 consonants. By

incorporating statistical information about allowable phoneme sequences in English, they

increased the overall phoneme recognition accuracy for words consisting of two or more

phonemes. This work marked the first use of statistical syntax (at the phoneme level) in

automatic speech recognition. An alternative to the use of a speech segmenter was the

concept of adopting a non-uniform time scale for aligning speech patterns. This concept

started to gain acceptance in the 1960‟s through the work of Tom Martin at RCA

Laboratories and Vintsyuk in the Soviet Union. Martin recognized the need to deal with the

temporal non-uniformity in repeated speech events and suggested a range of solutions,

including detection of utterance endpoints, which greatly enhanced the reliability of the

recognizer performance. Vintsyuk proposed the use of dynamic programming for time



6

alignment between two utterances in order to derive a meaningful assessment of their

similarity. His work, though largely unknown in the West, appears to have preceded that of

Sakoe and Chiba as well as others who proposed more formal methods, generally known as

dynamic time warping, in speech pattern matching. Since the late 1970‟s, mainly due to the

publication by Sakoe and Chiba, dynamic programming, in numerous variant forms

(including the Viterbi algorithm [19] which came from the communication theory

community), has become an indispensable technique in automatic speech recognition.

Advancement in technology

Figure shows a timeline of progress in speech recognition and understanding

technology over the past several decades. We see that in the 1960‟s we were able to

recognize small vocabularies (order of 10-100 words) of isolated words, based on simple

acoustic-phonetic properties of speech sounds. The key technologies that were developed

during this time frame were filter-bank analyses, simple time normalization methods, and the

beginnings of sophisticated dynamic programming methodologies. In the 1970‟s we were

able to recognize medium vocabularies (order of 100-1000 words) using simple template-

based, pattern recognition methods. The key technologies that were developed during this

period were the pattern recognition models, the introduction of LPC methods for spectral

representation, the pattern clustering methods for speaker-independent recognizers, and the

introduction of dynamic programming methods for solving connected word recognition

problems. In the 1980‟s we started to tackle large vocabulary (1000-unlimited number of

words) speech recognition problems based on statistical methods, with a wide range of

networks for handling language structures. The key technologies introduced during this

period were the hidden Markov model (HMM) and the stochastic language model, which

together enabled powerful new methods for handling virtually any continuous speech

recognition problem efficiently and with high performance. In the 1990‟s we were able to

build large vocabulary systems with unconstrained language models, and constrained task

syntax models for continuous speech recognition and understanding. The key technologies

developed during this period were the methods for stochastic language understanding,

statistical learning of acoustic and language models, and the introduction of finite state

transducer framework (and the FSM Library) and the methods for their determination and

minimization for efficient implementation of large vocabulary speech understanding systems.



7

Finally, in the last few years, we have seen the introduction of very large vocabulary systems

with full semantic models, integrated with text-to-speech (TTS) synthesis systems, and multi-

modal inputs (pointing, keyboards, mice, etc.). These systems enable spoken dialog systems

with a range of input and output modalities for ease-of-use and flexibility in handling adverse

environments where speech might not be as suitable as other input-output modalities. During

this period we have seen the emergence of highly natural concatenative speech synthesis

systems, the use of machine learning to improve both speech understanding and speech

dialogs, and the introduction of mixed-initiative dialog systems to enable user control when

necessary.

After nearly five decades of research, speech recognition technologies have finally

entered the marketplace, benefiting the users in a variety of ways. Throughout the course of

development of such systems, knowledge of speech production and perception was used in

establishing the technological foundation for the resulting speech recognizers. Major

advances, however, were brought about in the 1960‟s and 1970‟s via the introduction of

advanced speech representations based on LPC analysis and cepstral analysis methods, and in

the 1980‟s through the introduction of rigorous statistical methods based on hidden Markov

models. All of this came about because of significant research contributions from academia,

private industry and the government. As the technology continues to mature, it is clear that

many new applications will emerge and become part of our way of life – thereby taking full

advantage of machines that are partially able to mimic human speech capabilities.



8



9

3. METHODOLOGY OF THE PROJECT

The methodology of the project involves the following steps

1. Database collection

2. Decomposition of the speech signal

3. Feature vectors extraction

4. Developing a classifier

5. Training the classifier

6. Testing the classifier

Each of the section is discussed in detail below

3.1 Database collection

Database collection is the most important step in speech recognition. Only an

efficient database can yield a good speech recognition system. As we know different people

say words differently. This is due to the difference in the pitch, slang, pronunciation. In this

step the same word is recorded by different persons. All words are recorded at the same

frequency 16KHz. Collection of too much samples need not benefit the speech recognition.

Sometimes it can affect it adversely. So, right number of samples should be taken. The same

step is repeated for other words also.

3.2 Decomposition of speech signal

The next step is speech signal decomposition. For this we can use different

techniques like LPC, MFCC, STFT, wavelet transform. Over the past 10 years wavelet

transform is mostly used in speech recognition. Speech recognition systems generally carry



10

out some kind of classification/recognition based upon speech features which are usually

obtained via time-frequency representations such as Short Time Fourier Transforms (STFTs)

or Linear Predictive Coding (LPC) techniques. In some respects, these methods may not be

suitable for representing speech; they assume signal stationarity within a given time frame

and may therefore lack the ability to analyze localized events accurately. Furthermore, the

LPC approach assumes a particular linear (all-pole) model of speech production which

strictly speaking is not the case.

Other approaches based on Cohen‟s general class of time-frequency distributions

such as the Cone-Kernel and Choi-Williams methods have also found use in speech

recognition applications but have the drawback of introducing unwanted cross-terms into the

representation. The Wavelet Transform overcomes some of these limitations; it can provide a

constant-Q analysis of a given signal by projection onto a set of basic functions that are scale

variant with frequency. Each wavelet is a shifted scaled version of an original or mother

wavelet. These families are usually orthogonal to one another, important since this yields

computational efficiency and ease of numerical implementation. Other factors influencing the

choice of Wavelet Transforms over conventional methods include their ability to capture

localized features.

Tiling of time frequency plane via the wavelet transform



11

Wavelet Transform

The Wavelet transform provides the time-frequency representation. (There are other

transforms which give this information too, such as short time Fourier transform, Wigner

distributions, etc.)

Often times a particular spectral component occurring at any instant can be of

particular interest. In these cases it may be very beneficial to know the time intervals these

particular spectral components occur. For example, in EEGs, the latency of an event-related

potential is of particular interest (Event-related potential is the response of the brain to a

specific stimulus like flash-light, the latency of this response is the amount of time elapsed

between the onset of the stimulus and the response).

Wavelet transform is capable of providing the time and frequency information

simultaneously.

Wavelet transform can be applied to non-stationary signals. It concentrates into

small portions of the signal which can be considered as stationary. It has got a variable size

window unlike constant size window in STFT. WT gives us information about what band of

frequencies is there in a given interval of time.



12

There are two methodologies for speech decomposition using wavelet. Discrete

Wavelet Transform (DWT) and Wavelet Packet Decomposition (WPD). Out of the two DWT

is used in our project.

Discrete Wavelet Transform

The transform of a signal is just another form of representing the signal. It does

not change the information content present in the signal. For many signals, the low-frequency

part contains the most important part. It gives an identity to a signal. Consider the human

voice. If we remove the high-frequency components, the voice sounds different, but we can

still tell what‟s being said. In wavelet analysis, we often speak of approximations and details.

The approximations are the high- scale, low-frequency components of the signal. The details

are the low-scale, high frequency components. The DWT is defined by the following

equation:

Where ψ(t) is a time function with finite energy and fast decay called the mother wavelet.

The DWT analysis can be performed using a fast, pyramidal algorithm related to multi-rate

filter-banks. As a multi-rate filter-bank the DWT can be viewed as a constant Q filter-bank

with octave spacing between the centers of the filters. Each sub-band contains half the

samples of the neighboring higher frequency sub-band. In the pyramidal algorithm the signal

is analyzed at different frequency bands with different resolution by decomposing the signal

into a coarse approximation and detail information. The coarse approximation is then further

decomposed using the same wavelet decomposition step. This is achieved by successive

high-pass and low-pass filtering of the time domain signal and is defined by the following

equations:



13

Figure 1: Signal x[n] is passed through lowpass and highpass filters and it is down

sampled by 2

In the DWT, each level is calculated by passing the previous approximation

coefficients though a high and low pass filters. However, in the WPD, both the detail and

approximation coefficients are decomposed.

Figure 2: Decomposition Tree

The DWT is computed by successive low-pass and high-pass filtering of the

discrete time-domain signal as shown in figure 1 and 2. This is called the Mallat algorithm or

Mallat-tree decomposition.

The mother wavelet used is daubichies 4 type wavelet. It contains more number

of filters. Daubichies wavelets are the most popular wavelets. They represent the foundations

of wavelet signal processing and are used in numerous applications. These are also called

Maxflat wavelets as their frequency responses have maximum flatness at frequencies 0 and π



14

Daubechies wavelet of order 4

3.3 Feature vectors extraction

Feature extraction is the key for ASR, so that it is arguably the most important

component of designing an intelligent system based on speech/speaker recognition, since the

best classifier will perform poorly if the features are not chosen well. A feature extractor

should reduce the pattern vector (i.e., the original waveform) to a lower dimension, which

contains most of the useful information from the original vector.

The extracted wavelet coefficients provide a compact representation that

shows the energy distribution of the signal in time and frequency. In order to further

reduce the dimensionality of the extracted feature vectors, statistics over the set of the

wavelet coefficients are used. That way the statistical characteristics of the “texture” or the

“music surface” of the piece can be represented. For example the distribution of energy in

time and frequency for music is different from that of speech.

The following features are used in our system:

The mean of the absolute value of the coefficients in each sub-band. These features

provide information about the frequency distribution of the audio signal.

The standard deviation of the coefficients in each sub-band. These features provide

information about the amount of change of the frequency distribution.

Energy of each sub-band of the signal. These features provide information about the

energy of the each sub-band.

Kurtosis of each sub-band of the signal. These features measure whether the data are

peaked or flat relative to a normal distribution.



15

Skewness of each sub-band of the signals. These features are the measure of

symmetry or lack of symmetry.

These features are then combined into a hybrid feature and are fed to a classifier.

Features are combined using a matrix. All the features of one sample correspond to a column.

3.4 Developing a classifier

Generally, there are three usual methods in speech recognition: Dynamic Time

Warping (DTW), Hidden Markov Model (HMM) and Artificial Neural Networks (ANNs).

Dynamic time warping (DTW) is a technique that finds the optimal alignment

between two time series if one time series may be warped non-linearly by stretching or

shrinking it along its time axis. This warping between two time series can then be used to find

corresponding regions between the two time series or to determine the similarity between the

two time series.

In speech recognition Dynamic time warping is often used to determine if two

waveforms represent the same spoken phrase. This method is used for time adjustment of two

words and estimation their difference. In a speech waveform, the duration of each spoken

sound and the interval between sounds are permitted to vary, but the overall speech

waveforms must be similar. Main problem of this systems is little amount of learning words

high calculating rate and large memory requirement.

Hidden Markov Models are finite automates, having a given number of states;

passing from one state to another is made instantaneously at equally spaced time moments.

At every pass from one state to another, the system generates observations, two processes are

taking place: the transparent one, represented by the observations string (feature sequence),

and the hidden one, which cannot be observed, represented by the state string. Main point of

this method is timing sequence and comparing methods.

Nowadays, ANNs are utilized in wide ranges for their parallel distributed

processing, distributed memories, error stability, and pattern learning distinguishing ability.

The Complexity of all these systems increased when their generality rises. The biggest



16

restriction of two first methods is their low speed for searching and comparing in models. But

ANNs are faster, because output is resulted from multiplication of adjusted weights in present

input. At present TDNN (Time-Delay Neural Network) is widely used in speech recognition.

Neural Networks

A neural network (NN) is a massive processing system that consists of many

processing entities connected through links that represent the relationship between them. A

Multilayer Perceptron (MLP) network consists of an input layer, one or more hidden layers,

and an output layer. Each layer consists of multiple neurons. An artificial neuron is the

smallest unit that constitutes the artificial neural network. The actual computation and

processing of the neural network happens inside the neuron. In this work, we use an

architecture of the MLP networks which is the feed-forward network with back-propagation

training algorithm (FFBP). In this type of network, the input is presented to the network and

moves through the weights and nonlinear activation functions toward the output layer, and

the error is corrected in a backward direction using the well-known error back-propagation

correction algorithm. The FFBP is best suited for structural pattern recognition. In structural

pattern recognition tasks, there are N training examples, where each training example consists

of a pattern and a target class (x,y). These examples are assumed to be generated

independently according to the joint distribution P(x,y). A structural classifier is then defined

as a function h that performs the static mapping from patterns to target classes y=h(x). The

function h is usually produced by searching through a space of candidate classifiers and

returning the function h that performs well on the training examples during a learning

process. A neural network returns the function h in the form of a matrix of weights.



17

An Artificial Neuron

The number of neurons in each hidden layer has a direct impact on the

performance of the network during training as well as during operation. Having more

neurons than needed for a problem runs the network into an over fitting problem. Over fitting

problem is a situation whereby the network memorizes the training examples. Networks that

run into over fitting problem perform well on training examples and poorly on unseen

examples. Also having less number of neurons than needed for a problem causes the network

to run into under fitting problem. The under fitting problem happens when the network

architecture does not cope with the complexity of the problem in hand. The under fitting

problem results in an inadequate modeling and therefore poor performance of the network.

MLP Neural network architecture



18

The Backpropagation Algorithm

The backpropagation algorithm (Rumelhart and McClelland, 1986) is used in

layered feed-forward ANNs. This means that the artificial neurons are organized in layers,

and send their signals “forward”, and then the errors are propagated backwards. The network

receives inputs by neurons in the input layer, and the output of the network is given by the

neurons on an output layer. There may be one or more intermediate hidden layers. The

backpropagation algorithm uses supervised learning, which means that we provide the

algorithm with examples of the inputs and outputs we want the network to compute, and then

the error (difference between actual and expected results) is calculated. The idea of the

backpropagation algorithm is to reduce this error, until the ANN learns the training data. The

training begins with random weights, and the goal is to adjust them so that the error will be

minimal.

3.5 Training the classifier

After development the classifier has got 2 steps. Training and testing. In

training phase the features of the samples are fed as input to the ANN. The target is set. Then

the network is trained. The network will adjust its weights such that the target is achieved for

the given input. In this project we have used the function „tansig‟ and „logsig‟. So the output

should be bounded between 0 and 1. The output is given as .9 .1 .1……1 for 1st word. .1 .9

.1…….1 for 2nd

word and so on. The position of maximum value corresponds to the output.

3.6 Testing the classifier

The next phase is testing. The samples which are set aside for testing is given

to the classifier and the output is noted. If we don‟t get the desired output ,we reach the

required output by adjusting the number of neurons.



19

4. OBSERVATION

We recorded five Malayalam words “onnu”, ”randu”, ”naalu” , ”anju “ and

“aaru” .The words corresponds to Malayalam words for numerals 1,2,4 ,5 and 6 respectively.

The reason for specifically selecting these words was that,the project was intended to

implement a password system with numerals.

Malayalam Word Numeral

ഒ

1

2

4

5

6



20

20 samples for each word was recorded from different people and these samples were

then normalized by dividing their maximum values.Then they were decomposed using

wavelet transform technique upto eight levels since majority of the information about the

signal is present in the low frequency region.

In order to classify the signals an ANN is developed and trained by fixing outputs such

that

If the word is „onnu‟ then output will be .9 .1 .1 .1 .1

If the word is „randu‟ then output will be .1 .9 .1 .1 .1

If the word is „naalu‟ then output will be .1 .1 .9 .1 .1

If the word is „anju‟ then output will be .1 .1 .1 .9 .1

If the word is „aaru‟ then output will be .1 .1 .1 .1 .9

Out of 20 samples recorded,16 samples are used to train the ANN and the unused 4 samples are

used for test purpose.

Plots

Plot for word ‘onnu’



21

Plot for word ‘randu’

Plot for word ‘naalu’



22

Plot for word ‘anju’

Plot for word ‘aaru’



23

DWT Tree

The 8 level decomposition tree for a signal using DWT is shown in the

figure,which produces one approximation coefficient and eight detailed

coefficients



24

Decomposed waveforms for word ‘randu’



25

Decomposed waveforms for word ‘aaru’



26

5. TESTING AND RESULT

Testing with pre-recorded samples

Out of the 20 samples recorded for each word, 16 were used for training purpose.

We tested our program‟s accuracy with these 4 unused samples. A total of 20 samples were

tested ( 4 samples each for the 5 words) and the program yielded the right result for all 20

samples. Thus, we obtained 100% accuracy with pre- recorded samples.

Real-time testing:

For real-time testing, we took a sample using microphone and directly executed the

program using this sample. A total of 30 samples were tested, out of which 20 samples gave

the right result. This gives an accuracy of about 66% with real-time samples.



27

Change in efficiency by changing the parameters of the ANN were observed and

are plotted below

Plot 1: Accuracy with 2 layer feed forward network,Number of neurons in the first layer=15



28

Plot 2: Accuracy with 2 layer feed forward network ,Number of neurons in the first layer=20



29

Plot 3: Accuracy with 3 layer feed forward network,Number of neurons in the first

layer,N1=15& number of neurons in the second layer,N2=5



30

7. CONCLUSION

Speech recognition is one of the advanced areas. Many research works has been

taking place under this domain to implement new and enhanced approaches. During the

experiment we experienced the effectiveness of Daubechies4 mother wavelet in feature

extraction. In this experiment we have only used a limited number of samples. Increasing the

number of samples may give better feature and a good recognition result for Malayalam word

utterances. The performance of Neural Network with wavelet is appreciable. We have used

software with some limitations, if we increase the number of samples as well as the number

iterations (training), it can produce a good recognition result.

We also observed that, Neural Network is an effective tool which can be embedded

successfully with wavelet. The effectiveness of wavelet based feature extraction with other

classification methods like neuro-fuzzy and genetic algorithm techniques can be used to do

the same task.

From this study we could understand and experience the effectiveness of discrete

wavelet transform in feature extraction. Our recognition results under different kind of noise

and noisy conditions, show that choosing dyadic bandwidths have better performance than

choosing equal bandwidths in sub-band recombination. This result adapts to way which

human ear recognizes speech and shows a useful benefit of dyadic nature of multi-level

wavelet transform for sub-band speech recognition.

The wavelet transform is a more dominant technique for speech processing

than other previous techniques. ANN has proved to be the most successful classifier

compared to HMM.



31

8. REFERENCES

[1] Vimal Krishnan V.R, Athulya Jayakumar, Babu Anto.P, “Speech Recognition of Isolated Malayalam

Words Using Wavelet Features and Artificial Neural Network”, 4th

IEEE International Symposium on

Electronic Design, Test & Applications

[2] Lawrance Rabiner, Bing-Hwang Juang, “Fundamentals Speech Recognition”, Eaglewood Cliffs, NJ,

Prentice hall, 1993.

[3] Mallat Stephen, “A Wavelet Tour of Signal Processing”, San Dieago: Academic Press, 1999, ISBN

012466606.

[4] Mallat SA, “Theory for MuItiresolution Signal Decomposition: The Wavelet Representation”, IEEE

Transactions on Pattern Analysis Machine Intelligence. Vol. 31, pp 674-693, 1989.

[5] K.P. Soman, K.I. Ramachandran, “Insight into Wavelets from Theory to Practice”, Second Edition, PHI,

2005.

[6] Kadambe S., Srinivasan P. “Application of Adaptive Wavelets for Speech “, Optical Engineering 33(7),

pp. 2204-2211, July 1994.

[7] Stuart Russel, Peter Norvig, “Artificial Intelligence, A Modern Approach”, New Delhi: Prentice Hall of

India, 2005.

[8] S.N. Srinivasan, S. Sumathi, S.N. Deepa, “Introduction to Neural Networks using Matlab 6.0,” New Delhi,

Tata McGraw Hill, 2006.

[9] James A Freeman, David M Skapura, “Neural Networks Algorithm”. Application and Programming

Techniques, Pearson Education, 2006.


Owner

FYP Stamp

Speech recognition-using-wavelet-transform

Technology

Transcript of Speech recognition-using-wavelet-transform