VOICE RECOGNITION SYSTEM
Nikhil Gupta
Siddharth Rawat
Devanshu Arya
1. PROBLEM STATEMENT
To recognize a word spoken by the user
To display the correctly recognized word on a character LCD.
2. OVERVIEW
The function of the voice recognition system is to :
Create a dictionary of 3-5 words.
Compare a test word spoken by the user with the words stored in the
dictionary and display the word if a match is found.
3. Basic Block Diagram
Micro-phone
ATmega 32 microcontroller
LCD Display
4. STRUCTURE
The microphone analog circuit goes to the ADC of the MCU. The digitized
sampling of the word is passed through the digital Chebyshev filters (flash
programmed onto the MCU). Once the analysis is done, a fingerprint of the
word is created and stored in the memory.
While recording the stored fingerprints form the dictionary, against
which the fingerprint of the test word is compared using methods
such as Euclidean distance and correlation.
At the time of testing, once the comparisons between the test word
and the dictionary are made, the LCD which is connected to the MCU
displays the word spoken if it has a match in the dictionary.
The software tradeoff in this project is between the number of filters we can
implement and the maximum number of cycles we have to adhere to. The
more filters there are, the more accurate the speech recognition will be.
However, because each filter takes about 320 cycles and we could not
implement more than 2000 cycles, we had to trade off the accuracy of the
system and limit the number of filters to 5.
5. ALGORITHMS & TECHNIQUES
What we needed to know in this project is how to calculate the frequency to
sample speech based on the Nyquist Rate Theorem. According to this
theorem, the sampling rate must be atleast twice the maximum frequency of
the sample input.
We had to learn about Chebyshev filters to determine the cutoff frequencies
to build the digital filters for human voice.
As for the analysis part of the speech, we needed to know how to calculate
euclidean distance and correlation.
Tor's speech recognition algorithm: A complex algorithm for voice
recognition, which uses the above-mentioned concepts and algorithms.
6. SPEECH RECOGNITION ALGORITHM
According to the algorithm used in our project, every 250th sample of the
input word is analyzed by the digital filters. Each of the five filters then
returns some numerical value which is stored in an array.
The process is repeated till the 2000th sample is analyzed, resulting in an array
of 40 elements. This is the fingerprint of the spoken word.
7. CIRCUIT DIAGRAM FOR MICROPHONE CIRCUIT
8. ANALOG CIRCUIT
The analog circuit consisted of a microphone, a high-pass filter and a series of
op-amps.
The microphone output voltage was of the order of a few millivolts. We had
to use two op-amps to get a voltage output of about 5 volts.
The microphone was directly connected to a capacitor(high-pass filter) to
remove signals with frequency less than 50Hz.
The output of the second op-amp was directly connected to the ADC of the
ATmega32.
9. PRACTICAL TESTING
We started by testing for a single word. However, we faced several problems
and had to make several changes to the code. We also tried different words,
and finally achieved 100 per cent accuracy for two words. The words used in
this case were BYE and ANGRY.
We then proceeded to perfect the voice recognition system for atleast four
words. We attempted several different combination of words such as HELLO,
LEFT, CLOSE, ON, OFF, AHEAD and many more, but achieved success with BYE,
GO, RIGHT and DOWN.
10. DICTIONARY SIZE AND CORRELATION
An observation that we made while testing was there was a relation between
the number of words of the dictionary and the value of minimum correlation.
The optimum value of minimum correlation for single word testing was 0.1.
This value reduced to 0.0001 when the dictionary size was increased to four
words.
11. TASKS COMPLETED
Voice recognition achieved for two words with 100 per cent accuracy.
Voice recognition for four words with accuracy close to 80 per cent.
12. FUTURE PROSPECT
A complete voice recorder and recognizer
Voice-controlled robotic car
Voice-controlled security system
13. REFERENCES
Tor's Speech Recognition Reference Code:
[Link]
Bruce's Digital Filters:
Code[Link]
[Link]
2/mysite/[Link]
[Link]
2/mysite/Clabwdf.c
[Link]/gadgets/high-tech-gadgets/[Link]