SAPI BASED APPLIANCES

CONTROL (VACS)

BS (Electronics Engineering)
Project Proposal
Batch-2007

Compiled by
Syed Mehtab Hasan EE-2007-233
Mohammad Junaid Zafar EE-2007-190
Syed Faraz Shah EE-2007-240
Mohammad Wasim EE-2007-215
Mohammad Owais EE-2007-529
Mohammad Ajaz Qurashi EE-2007-525

Internal Advisor Jawad Bhatti

Sir Syed University of Engineering & Technology

Transmitting, receiving, and processing speech is one of the largest, if not the

largest, area of communication technology. Because my emphasis is in this field of

electronics engineering, I felt I could gain an edge and much needed experience by doing

a project on speech recognition. This project will give further understanding of related

areas, such as speech synthesis, speaker recognition, and large vocabulary speech

recognition (for word processing). Thus, I find the project very interesting, applicable,

and practical.

Objectives:

The VACS will recognize simple spoken commands, such as “lights on,” and

respond accordingly; that is, it will send a wireless signal to a module connected to the

“lights,” telling the module to turn them “on.”

This same scenerio can be used for many other household systems—electric door

locks, CD changers, ceiling fans—anything that has a switch or is remote controlled.

Even old televisions that do not have a remote control—The power, volume, and channel

buttons can be “tapped” into and hooked up to remote modules. Individual lights can be

“named” and assigned a module. All additional modules can be controlled by one central

speech recognition unit.

Customer benefits:

• Ends hassle of finding a light switch in the dark, just say the light’s name

• Complete control over separate lights or appliances

• Control words are completely reprogrammable

• Add more modules to the existing control unit, anytime

• Want to make sure you turned off the stove? Call from New Zealand and leave a

message on your answering machine: “stove…off.”

Product features:

• Stand alone system—Entire system consists of control unit and multiple modules,

no PC connection necessary

• Portable control unit—Place in any room for best sound reception

• Different modules can be used for different applications—Control a 220V outlet

or a 5V power switch

• Continuous time signal analysis

• Sensitive receiver “hears” command from far away

II. DESIGN

Block Diagram:

Signal Microphone Filter A/D

Remote
Module
Block Description:

Input Signal—This incoming audible signal includes both speech and noise. It

will be processed continually as the digital signal processor searches for a recognizable

command word.

Microphone—The microphone will receive the sound waves and convert them to

a continuous (analog) variable voltage signal. The microphone’s sensitivity will

determine how far away the person can be to send a successful command.

Filter—Only frequencies between 300 Hz and 3 kHz (typical vocal range) will

pass through the filter, eliminating most noise. The filtered analog signal will then be

sent to the analog to digital converter.

A/D Converter—Here, the analog signal is converted into a digital pulse code

modulated (PCM) signal. Since a single command word must be recognized amid

background noise and similar-sounding words, the digital encoding must be precise. A

typical sampling rate used for vocal-PCM signals is 8000 samples/sec (above the Nyquist

rate of 6000 samples/sec). Twelve bits/sample is sufficient for linear quantization. Thus,

the A/D converter must output 12 kB/sec, continually.

Microprocessor—While the system is running, this digital signal must be

searched, in continuous time, for a recognizable command. This is the job of the

microprocessor. The microprocessor will break the signal down into discrete time

segments and compare them with the stored values in memory. If a match is found, the

microprocessor will send a signal to the transmitter, telling it, in turn, to transmit a
specific data sequence to the modules. The microprocessor will also allow the control

words to be stored in memory.

Wireless Transmitter—The transmitter outputs an eight bit signal to the modules,

specifying the module corresponding to the control word, and how the module should

respond. The transmitter must have long range, sending the signal throughout the house,

yet transmit the data at a frequency that will not cause interference with radio, television,

or phone signals.

Remote module—“On” and “off” of the appliance is directly controlled by the

module. Every module will receive the signal sent from the transmitter, but only one

module will respond to the signal, turning its appliance ‘on’ or ‘off.’

Memory—The memory is used for storage of the control words. The

microprocessor will control reading and writing from/to the memory.

Digital Signal Processor—The DSP board already has the filter, A/D converter,

microprocessor, memory, and transmitter built in. The board will need to be programmed

to utilize all of these components in the correct manner.

Performance Requirement:

Each of the components listed above greatly affect the overall quality of the

VACS. The main feature affecting performance will be the DSP’s capability to filter out

noise from the desired signal. With a sensitive microphone, any noise within the house

will be received also, making it much harder for the DSP to decipher the command word.

However, the faster the DSP’s processing capabilities, the better the chance of the

incoming digital stream to be recognized, since the processor can be programmed to

section, average, and process the stream with greater accuracy.

Testing procedures:

Much of the research and analysis for voice recognition will be done using

Matlab, since it offers a visual representation of audio signals, while the DSP does not.

Once the best method for recognizing a command word is found, the DSP will be

programmed to perform that method. Numerous methods currently exist for processing

speech, utilizing different ways of sectioning, averaging, and comparing the digital

stream. Matlab, with its ability to plot and compare sound waves, will be the fastest and

easiest way to experiment with these methods.

Since the main components are already contained on the DSP board, a large part

of this project will be programming and debugging the code.

Other hardware research will be done on the transmitter/receiver pair. It will be

necessary to find the range and reliability of the signal. This can be done using simple

methods. For instance, send a continuous signal from the transmitter telling the receiver

to turn a light on and off repeatedly. Then, walk the receiver and light away from the

transmitter. Where the signal is becomes too weak to read all the time, the light will

blink slower, or stop blinking. This is the true range of the receiver/transmitter pair.
Tolerance Analysis:

Obviously, a key part of the entire system is comparing the incoming data stream

with the stored control words in memory. The problem is that when a control word

(“lights”) is uttered, converted, and filtered, it will not have the exact same digital

representation as the control word in memory. Therefore, a certain tolerance must be

allowed for matching words. But, if too much tolerance is allowed, words that sound

similar to the control word (“life” compared to “light”) will also activate the system.

The research done on Matlab will conclude how much tolerance can be allowed.