|
Design 1
Utilizing Voice Extreme
|
Prototype
The cost to produce this prototype exceeds
our $50 budget.
Solution is to work together as a class
and design one or more clapper-type eBlock in parallel
Cost of development kit (includes 1
module) is $130
Additional prototype modules $60
Total costs estimate ~$150 after
the inclusion of the costs of the PIC, microphone, power regulator,
resistors, capacitors, LEDs, etc.
 in
individual groups
as a class
Mass production
VE IC die < $1.50 according to
this Sensory
PDF in 100k quantities
VE IC = $15-20 in low quantities
2MB Flash Memory WINBOND 29C020 ~ $6 dollars in low quantities
Design custom module to lower costs
Total costs estimate of a a VE based
eBlock if producing 100,000 is roughly 8 to 10 dollars.
Producing any quantity less then 1000 will dramatically increase unit
cost since the VE IC and Flash in low quantities are costly as shown
above.
/
Depends on target price of this eBlock |
Prototype
The minimum which is needed to get a
functioning eBlock which meets our requirements would required the VE
module, PIC, microphone and battery. All these components fits
inside the current eBlock housing.
The size of the module is 4.255 cm by
4.325 cm. The current eBlock housing is 7.5 cm by 7 cm. A
small board with the PIC, microphone and the VE module will easily fit
inside along with the 9V battery in the eBlock casing.
Adding a speaker to provide feedback to
the user will also fit inside the eBlock.
Mass production
Since the prototype will fit in the
current housing, the volume version will most certainly fit.
Designing a custom board that connects the VE IC, memory, PIC will
reduce the space required further.
|
Prototype
satisfies
power constraint VE module has
low power requirements
Vdd = 2.85V - 3.3V
Idd = 26mA @ 3 V
If this design is left on using the
typical 3V at 26mA all day long, a 9V battery would not be able to
sufficiently power it. It would only be powered for about a
1/4th of a day.
The low power mode which exist on the
VE must be utilized to make this design meet power constraints.
Low power mode uses 5um.
I estimate the probability that we get
the VE module's low power mode operational at 90%
Volume version will most consume
similar levels of power since the main culprits are the VE IC and
flash memory. With the power down mode implemented, this should
not be a problem.
|
From a developer stand-point, highly
flexibility since application consists of a program file plus any data
files it needs, linked together into a binary image file that can be
downloaded to the 2MB flash memory. A quick change of the
program and a quick download to the flash provides excellent
flexibility.
Additional functionality can be added on
easily and the prototype can evolve to contain more features.
Documentation PDFs are well done.
PDFs exist for module, kit, starter manual. Tutorials which
accompany development kit demo the many possibilities that we can
implement.
Features include the ability to recognize
speech from everyone (speaker-independent) through the use of weight
files. These must be purchased from Sensory.
The VE based eBlock can also be
programmed to be (speaker-dependent), meaning it only recognizes the
commands of a certain individual.
Continuous Listening
Word Spotting
|
Processes speech in real-time. Has
the ability to recognizes speech from almost anyone as well as the
ability to process speech from particular individuals who have their
speech programmed into VE. Confidence of recognition can be
adjusted in code. Has the ability to do continuous listening
(listening for a word with .25 seconds of silence before and after the
word) and word spotting (the ability to detect words embedded in the
middle of sentences)
|
Maintainability for this design is very
good. The documentation for the VE is excellent and the PIC
documentation as well as the current eBlock code is well written.
Adjusting the programming for the VE as well as the PIC should not be
a problem for someone that did not design the system if he or she
spends some time reading the documentation.
|
Satisfies time-to-prototype
Design 1 satisfies the time to
prototype window. Assuming we have 7-8 weeks till the prototype
is due, (From the date we received dev. kit), a breakdown of the weeks
is as follows.
Week 1 through week 2 includes setup of
dev. kit, reading VE programmers guide as well as tutorials. As
well as getting acquainted with the VE hardware and software,
specifications for the prototype must be hammered out at this time to
so design phase will know exactly what to construct.
Week 3 and 4 will revolve on actual
design and will focus on programming the VE as well as the PIC for our
specific purpose.
Week 5 involves packaging the prototype
into a presentable package. In our case, making it fit into the
eBlock housing.
Week 6-8, test the prototype as well as
final documentation. This window allows time for modifications
and changes if something is wrong.
Depending on features desired, this
design can meet the time-to-prototype window of 8 weeks as shown above
with 100% confidence.
|
Time to market issues
After the prototype has shown that this
design is feasible, I estimate time to market will be around another
3-4 months with a confidence of 50%.
A custom board with the VE IC, memory,
PIC must be developed. Further design and test will be needed.
A way must be found to program many
flash memories with the same code since this is where the VE program
and data is stored. The VE IC only interprets what is on the
flash. / |
Probably the most elegant design from the
user's perspective. Works right out of the box recognizing the
pre-determined words by any speaker. No fuss for the user to
train and setup. The ability to just connect to other eBlocks
right out of the box is appealing.
|
|
Design 2
Utilizing Voice Direct II |
Prototype
The cost to produce the
prototype exceeds the $50 budget, but significantly less then that of
VE above
Cost of development kit
(includes 1 module) is $50
Have not found additional
prototype modules for sale
Total cost estimate ~$70
/
in individual groups (2 people maybe? Cost
slightly exceeds that of budget )
as a class
Mass production
VD II IC die <
$1.50 according to this Sensory
PDF in 100k quantity
VD II IC = $
unknown in low quantities
Design custom module to
reduce costs
/
Depends on target price
of this eBlock |
VD II module, PIC,
speaker, buttons, microphone, battery fits inside eBlock. The
size of the module is similar to that of the VE module, around 4cm by
4cm.
Documentation PDF scan
is lacking. Missing page 7-8. Necessary information such as what
some of the pins are for on module are undocumented. Creating an
eBlock using the module might be difficult as a result of this.
It appears the VD II kit
was designed to be easily put into a design with the motherboard that
comes with the kit. The only problem with this is that it
exceeds our size constraint.
The motherboard is
similar in size to that of the VE motherboard measuring approximately
14cm by 7 cm, obviously not fitting in the eBlock housing.
|
May or may not satisfies power constraint
VD II module has low power requirements.
According to Sensor's site, the VDII has low a low power mode.
From looking at the documentation, it
appears the motherboard can be connected to and run off a 9V battery
but the papers do not specify how long the 9V battery would last.
Vdd = 3.3V typical
To drive an output pin low, 2mA
If this design were to be left on all
day with it continuously processing speech, it would last about 3-4
days on a 9V battery without use of the low power feature. The
low power feature doesn't seem to be well documented so that's another
problem.
/ |
eBlock would have to be trained by user
before use. Train by using buttons on eBlock.
Speaker-dependent
Documentation PDF scan is lacking.
Missing page 7-8. Necessary information such as what some of the
pins are for on module are undocumented. Creating an eBlock
using the module might be difficult as a result of this.
It appears the VD II kit was designed
to be easily put into a design with the motherboard that comes with
the kit. The only problem with this is that it exceeds our size
constraint.
Continuous Listening
Word Spotting
|
Processes speech in real time but limited
to processing speech that has been recorded into the block by user.
Can train phrases that are up to 2.5 seconds in length to be
recognized. The VDII claims 99% accuracy when trained properly.
This design would have the ability to store up to 16 commands or
phrases. Has the ability to do continuous listening (listening
for a word with .25 seconds of silence before and after the word) and
word spotting (the ability to detect words embedded in the middle of
sentences)
|
Maintainability for this design ranges
from below average to average. The documentation for the PIC as
well as the existing eBlock code is good but the documentation for the
Voice Direct II is poor. The PDF pages are in an incorrect order
with pages 7-8 missing. Those who look to maintain the system if
flaws are found will have a difficult time especially if they are not
the ones who designed the system. The only solution to this is
to provide precise documentation of the design and what was learned
during design in written form to the maintainers. If this is not
done, the ability to diagnose problems in this design as well as
making changes accordingly will be exceedingly difficult.
/ |
May or may not satisfy time to prototype.
The chance a working prototype will be finished by the specified date
is 80%. Assuming we have 7-8 weeks till the prototype is due,
(From the date we received dev. kit), a breakdown of the weeks is as
follows. Week 1, learn how to use
the VDII module and motherboard. Week 2-4 would entail find a
way to connect buttons, the microphone, and speaker to the VDII
module. Week 5-6 would entail interfacing an output pin form the
VDII module to the PIC followed by packaging the components into the
eBlock casing. Week 7-8 are reserved for final testing and
documentation.
|
Time to market issues.
Given that the prototype is feasible, I
estimate time to market will be 3 months with 80% confidence.
Unlike design 1, there is no worry of programming a large number of
flash memory.
|
Some potential users of a voice
recognition enabled eBlock could be turned off by the need to train
the eBlock before using it. There are those who actually might
prefer the training thus allowing the eBlock to only recognize a
particular user. A market survey probably has to be conducted to
determine what people prefer. If user's like the ability to have
an eBlock which only responds to their trained words, then this design
has an advantage over the others.
/ |
|
Design 3
Clapper-Type |
Not too much difference in price from
prototype and volume production.
Stays within $50 dollar prototype budget easily.
The required parts are quite common,
easily accessible and inexpensive such as op-amp, timers, gates, resistors and
capacitors.
Several designs exist. All of the
designs are low costs.
Total cost estimate ~ $10-15 for
prototype assuming we already have the PIC programmer.
Estimate for volume production version
is less then 10 dollars.
|
Definitely fits inside eBlock. PIC,
op-amp, resistor and capacitors on a small board plus 9V battery.
Existing eBlocks have similar components so this design will fit
inside an eBlock casing with 100% confidence.
|
From looking at other clapper-type
schematics on internet, this design will meet power constraint.
Must find a way to have the PIC power down to conserve power.
|
Does not satisfy the fundamental
requirement to be activated by voice. Operates by recognizing
claps.
Design does not offer flexibility to
evolve the product to have more features and capabilities.
|
Does not process speech. Processes
the amplitude from microphone and compares it to a reference
amplitude. If amplitude is reached at any given moment which can
be caused by any sound, then block outputs a yes.
/
|
Maintainability for this design is good since it is relatively simple
with code only existing only for the PIC. The clapper detector
circuit is completely done with hardware. The PIC and existing
eBlock code is well documented.
|
Definitely satisfies time-to-prototype
with 100% confidence. With the given 10 weeks to prototype,
there is a high confidence that this design will succeed with excess
time left over thus the reason for the development of parallel
projects (the VE based eBlock and poor man's)
|
Time to market issues. No real
issues to contend with. I estimate a time to market of another
two months after the completion of the prototype with most of that
dedicated to manufacturing time with 90% confidence.
|
Very intuitive for the user to
understand how to operate but much less elegant then the VE based
design. Works right out of the box. Able to connect to
other eBlocks right away without training.
|
|
Design 4 Custom
Design |
Should stay within $50 dollar prototype
budget assuming we have the PIC programmer. Additional ICs required such as those
to store the voice (memory in terms of EEPROM) as well as possibly A/D converters,
multipliers, amps should be
relatively inexpensive.
|
Definitely fits inside eBlock. Even
with the addition of other inexpensive ICs alongside the PIC, the
design and a 9V battery should will fit inside an eBlock.
|
Meets power constraint since using only a
PIC and possibly some A/D converter components. Additional
components may be necessary which would increase power consumption.
Developing a sleep mode is critical.
If design not use PIC, the design may or
may not meet power constraint. Other controllers may be consume
more power but most microcontrollers are built with power in mine so
this should not be a major issue.
|
Features would depend heavily on what we
decide to implement.
No matter what features are
implemented, some type of algorithm will need to be implemented to
handle the speech processing. Data storage will be a factor as well as
an “image” of the voice or speech pattern that is desired to be
recognized must be stored in the eBlock. This “image” will be needed
as a reference to be used for comparison to the sound being brought in
by the microphone. The sound being brought in by the microphone must
also be stored into memory since the PIC will most likely not be able
to compare the input to that of the reference speech pattern in
real-time. The time to process the comparison of the two will most
likely be excessive causing the time before the block outputs a yes or
no to be much delayed after the initial talking by the user. FFT
and the concept of windowing is of great importance.
Different types of microphones will
produce different speech signals for the same spoken word. Two types
exist, condenser and dynamic.
A design would have to be designed around a particular type of
microphone.
|
Designed to processes speech.
Depending on what we type of processor we use, this may or may not be
in real-time. Since we would be developing this, speaker
independent reorganization as well as continuous listening are the
only features we would include to simply the task. With
continuous listening, the design would only process words that were
preceded and followed by silence.
|
Maintainability will depend heavily on the
documentation habits and comments written alongside the code by the
designers of the system. If a good job is done with
documentation, I do not see a problem with maintaining and improving
the system for later revisions. If a poor job is done with
documentation, those who maintain the system will have a difficult
time.
/
|
Severely exceeds time-to-prototype.
Time must be spent researching and understanding concepts of voice
recognition. Time must be spent understanding what the PIC
can do as well as it's strong and weak points. Research into
alternative processors or FPGAs also adds time. This could aid in
understanding how to implement a voice recognition concept into actual
software or hardware. I estimate a more realistic goal of 12
months for time-to-prototype for this design. This time can be
reduced by bringing in someone who has knowledge about speech
processing and sharing his or her knowledge first hand.
|
Time to market issues. Most of the
time will be manufacturing related. The largest hurdles for this
design exist in the time to prototype phase. If a working
prototype is achieved, the time to market from the date the prototype
is finished will be insignificant compared to that of the time to
prototype. As with the other designs, I estimate another 3
months.
|
This design should provide the same
usability as design 1 utilizing the VE. Works right out of the
box.
|