An Najah National University facility of engineering department of Electrical Engineering

6.6 Conclusion

The project has not met our expectations fully, as we initially specified that the system would be able to recognize a sentence as a command. But we are more than happy that it is able to recognize a word as the command by more than 70%-80% of the time, depending on the command. There is a training procedure that needs to be implemented, which is an added feature to increase the accuracy of the program. However, the system can still be used without training but with much lower accuracy.

There are two hardest parts in our speech recognition project. One is for filter design, the other is fingerprints analysis. The shortcoming for filter is its frequency spectrum resolution is coarse and can't tell the difference in its band. So we have to select some distinct words as our codes. FFT is a good candidate for filter design and also for fingerprints analysis,.

Another problem is when a tester spoke the same word, even if there is a tiny difference when he spoke, the fingerprint changed a lot. We didn’t solve this problem until now. But we think if we increase frequency resolutions, maybe it will be helpful.

Actually, we have a big problem during the testing. We found the fingerprint of the same word will change a lot even if his pronunciation changes a little. So tried to record the same word for 20 times and get the average of the fingerprints. But we can't calculate their average value directly because their amplitude is quite different. So we use linear regression method i.e., try to normalize the every training sample to equivalent level then get their arithmetic average.

The program was able to recognize five words, but sometimes it would become confused and match the incorrect word if the word that was spoken varied too much from the word stored in the dictionary. As a rough estimate the program recognized the correct word about 70% of the time a valid word was spoken. The program achieved success using some voices, and with sufficient practice a person could say the same word with a small enough variation for the program to recognize the spoken word most of the time. For the general person though the recognition program would have a much lower percentage of success. Also the words in the dictionary are words spoken by only one person. If someone else said the same words it is unlikely the program would recognize the correct word most of the time, if at all.

For safety an testing we made sure the PWM signals sent to the car were as close to neutral as possible, while still letting the move go forward and backward. We did this to prevent the car from going out of control and potentially hurting others. Our project did not use any RF signals and the board we used ran just off of a battery so there were no physical connections to anything involving other people’s projects. Also the only pins switching state were the pins for the PWM, which were mostly covered by wire.

Using humanoid approach was not be able to our applications, and simple statistical was more robust and more accurate. This conclusion will not remain valid if number of voice commands increased; because statistical approaches fail to work find thresholds to separate between values coming from each command.

Books ;

1. 
   Rpdman,Rebert “Computer Speech Technology” 1999,Boston Pub .
   
2. 
   Walter A.tribel ,Avtar “the 8088 and 8086 Microprocessor , interfacing”2000,prentice hall .inc .
   
3. 
   Stephen "Theory of Filter Amplifiers", Wireless Engineer (also called Experimental Wireless and the Wireless Engineer), vol. 7, 1930, and pp. 536-541.
   
4. 
   Robin R.Murrhy , “introduction to Al Robotics”,2000 press Cambridge .
   
5. 
   Stephen J.Chapman, “Electric Machinery Fundamentals” 1994 4^th edition mcgraw-hill.
   

1. 
   www.wikibidia.com
   
2. 
   www.microchip.com
   
3. 
   www.Mathworks.com
   

if sff (i-j)< sff (i) & sff(i+k)

sff(i)=sff(i);

else

sff(i)=0;

end

end

end

end





for i= 1:2500

if sff(i)<0.05

sff(i)=0;

end

end



n=sff(1:2000);

[c ns]=sort (n);

ns=flipud (ns);



% The Signals Database

x1=1;

while ns(x1)<2000

x11=x1;

x1=x1+1;

end

qw=ns(1:x11);



if x11>=3

q=ns(1:3);

else

fprintf('Record again')

end

q=sort (q)



if z==1

i1=q;

go=sf;

end



if z==2

i2=q;

st=sf;

end



if z==3

i3=q;

ba=sf;

end



if z==4

i4=q;

le=sf;

end



if z==5

i5=q;

rig=sf;

end

z=z+1;

end



ii =[i1 i2 i3 i4 i5]

Speech Recognition

Fs=44100;

fprintf('\n recod now\n')

s = wavrecord(2*Fs,Fs,'double');

t=0:1/(Fs):1.99999;





% The Butterworth Filter

Wp = [150 8450]/11025 ; Ws = [100 9450]/11025; Rp = 0.8; Rs = 30.8;

[n,Wn] = buttord(Wp,Ws,Rp,Rs);

[b,a] = butter(n,Wn);

sf=filter(b,a,s);



sf =sf/max(abs(sf));

wavplay (s,Fs);



%Spectral Analysis

[B,f] = specgram(sf,Fs,Fs);

sff=B.*conj(B);

sff(1:10)=0;

sff=sff/max(sff);









for i=101:2500

for j=1:100

for z=1:100

if sff (i-j)< sff (i) & sff(i+z)

sff(i)=sff(i);

else

sff(i)=0;

end

end

end

end





for i= 1:2500

if sff(i)<0.05

sff(i)=0;

end

end



n=sff(1:2000);





%hold on



[c ns]=sort (n);

ns=flipud (ns);



x1=1;

while ns(x1)<2000

x11=x1;

x1=x1+1;

end

qw=ns(1:x11);



if x11>=3

q=ns(1:3);

else

fprintf('recod again')

end

q=sort (q);

y=q



% Finding the Signal

x1 =norm(y-i1)/3;

x2=norm(y-i2)/3;

x3=norm(y-i3)/3;

x4=norm(y-i4)/3;

x5=norm(y-i5)/3;

x=[x1 x2 x3 x4 x5]

[s I]=min(x);

% Recognaize The Word

if I==1

fprintf('Go\n')

wavplay (go,Fs);

output=1;

elseif I==2

fprintf('Stop\n')

wavplay (st,Fs);

output=0;

elseif I==3

fprintf('back\n')

wavplay (ba,Fs);

output=2;

elseif I==4

fprintf('left\n')

wavplay (le,Fs);

output=5;

elseif I==5

fprintf('right\n')

wavplay (rig,Fs);

output=9;



end

plot (t,sf)

xlabel ('time (s)')

ylabel ('Amplitude')

Directory: sites -> eng.najah.edu -> files
sites -> Glossary for Chapter 1 Algorithm
sites -> North Carolina Inclusion Initiative Mapping Where Children with ieps are Being Served Purpose
sites -> Northern England’s set-jetting locations
sites -> Physical custody of 1033 program property accountibility form statement of Physical Custody: By signing for the below 1033 property I am a Law Enforcement Officer of the aforementioned Law Enforcement Agency
sites -> Nstructions for Acquiring Excess Equipment online, through the 1033 Program
sites -> Memorandum of agreement
files -> Final project 1 Report

Download 303.2 Kb.

Share with your friends: