It's beyond my reach right now, but have any of you guys done it?··They have·voice activation cars now, but has it been done·on a robotic platform?· Just curious.··· -SciTech02.
· I haven't done it, but a LONG time ago Steve Ciarcia did an article on voice recognition that used a bunch of active filters. And I believe there are a couple cheap voice recognition chips on the market. I would probably look at those.
Do you get the 800 channel comcast? Watch more Science Channel. It is my 110. There are always fun and cool robotic shows on.··
Yes. I recently saw something on this neat robot called Coyote. It used advanced vision to detect hand gestures, and strong sound systems to hear. It was amazingly advanced. The guys sweep his arm to coyote's right, and says" Coyote; go left". The robot then says in a robot voice:"You pointed one way and told me to go another."-(It was funny to see this reaction) the the guy said"Coyote. Go to my left."The robot amazingly did so
I'm not sure if it is hackable, but hear is a super-ultra mindsttorm bot coming out september that can reconized voice patterns. http://shop.lego.com/product.asp?p=8527&cn=55&d=11&t=5
Vioce Recognition depends on three things:
Patterns
Amplitude of Sund Waves
Frequency
Interesting.· By the way,·can you get·any simple sound-reacting tools for the boe bot?· Like when it hears a lound noise, it will detect it.· Other really simple robots (Ones that cost $20.00 or less) can do it, can you get somthing that makes·the boe bot do that?·
And back on the voice recognition chips, How would you intergrate it to a BS2?· Can you just conect them to the breadboard, or would·you have to completly take it apart and do it?· Then, how would you program it?·
Or better yet, what if I just got a Text-to-speach board and a microphone/speaker, then did some sort of PULSIN command.· It's probably way more complicated then that, but could it be done?··· -SciTech02
Sensory, Incorporated makes an inexpensive ($49) module among other products, called Voice Direct II, that though I haven't worked with - I have been told is quite easy to use and does a good job. Take a look here:
There are several methods one could use to process speach.· One that I have heard of that seems like it might be fun to implement, breaks the sounds down into a phonetic alphabet consisting of: · vowels diphthongs voiced constants unvoiced constants special phonemes · The basic idea is to establish frequency matching relationships for each sound in each group and specific pattern matching to a word database. · For example the word "Bird" would phonetically look something like· "B-ER-D" where the stored pattern would be key tones within each phoneme.· ·
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔ Beau Schwabe
IC Layout Engineer
Parallax, Inc.
One of my graduate foci was digital signal processing, with empahsis in biological signal processing, and took a few classes on voice recognition. The first issue that needs to be solved is decomposition of phonemes (IOW you must break the spoken word/phrase into individual pieces). Once you have done that you are then faced with the issue of recognizing phonemes. The voiced part of a phoneme isn't that difficult (vowel sounds and voiced consonants like the tail sound of the word "are") if there is a large enough steady state window (a region of repeated waveform of roughly the same amplitude). This part can be done using analog circuitry if you are skilled enough. Unvoiced areas (consonants) are much harder to recognise because they are typically fleeting and are marked by constantly changing waveform. Analysis of unvoiced sections is typically done using digital signal processing, neural networks or other pattern matching methods. Something that also has to be taken into consideration is whether the system is speaker dependant or independant (recognizes a single person's voice or anyone's voice). Speaker dependent is the easier way to go; since pattern matching is frequently employed on unvoiced consonants, with a single speaker, you can pattern match entire words or phrases. Speaker independant systems cannot pattern match entire words.
Pattern matching is not that easy of a field, it was a graduate level course for good reason, the statistics and other mathmatics involved (pattern matching methods) are not that easy to follow without a good background in mathmatics. In the course we wrote a program for mine detection using ground penetrating radar which took the reflected signal, passed it through a 26 bin FFT of radio frequencies, then comapred the 26 dimensional output vector to a sample set of 512 known vectors (256 vector signatures of mined areas and 256 vector signatures of unmined areas). Only about half of our class got better than a 50% (pure chance) guess rate (amusingly one student got it mostly right, but then accidentaly inverted the output boolean, so it was right 11% of the time).
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
6+6=10 (Long live the duodecimal system)
Post Edited (Paul Baker) : 4/4/2006 6:13:30 PM GMT
Wow!· I didn't know that there was that much stuff avaible!· I'm leaning to the voice recognition chips, they seem to work well, look easy to conect to your robot, and have a great price.· Thanks for your help so far.· One more thing, I knew voice recognition was advanced, but you have to go to graduate school to learn it, wow.··· -SciTech02
My differential equations professor use to say in response to complicated questions he was asked "You didn't pay enough tuition to get an answer to that question, you'll need to go enroll at MIT to get an answer." Of course he would answer the question after the remark. It was his way of saying "You can learn something anywhere, paying a huge amount on education doesn't provide access to a hidden vault of knowledge." The same is true with learning something within or without college. The material is there for anyone to learn, but most people don't aquire the nessesary background knowledge until thier well within college (though nothing prevents you from doing so under your own volition).
For example, here is a fairly good write up of Bayes classifier, the first pattern recognition algorithm taught to students studying pattern recognition. But as you see if you don't already understand statistics and some set theory, the write up won't make sense to you.
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
6+6=10 (Long live the duodecimal system)
I have an obsolete Archer VCP200 'speaker-independent word recognizer' chip, which I bought on eBay. This was the chip Radio Shack sold. It recognizes five motion commends: go, stop, reverse, 'turn right' and 'left turn'. The commands switch active low outputs which can sink up to 10mA. It is 5 volt powered and looks ideal to work with the Stamp. I hven't had a chance to try it out yet.
I suggest that you search for this chip on eBay and other sources of old ICs.
Nice.· Can you make it do your own commands?· For example, this is what I wanted to do.··I was going to make it similar to the LCD screen in that you can save multiple programs and then say; "Do this program" or "Do that program" or "Stop" and so on.· Could I make it do that?··· -SciTech02.
Unfortunately not. The Radio Shack voice module was very limited to 4-5 voice commands (or another mode with Yes or No as commands). It·should have·a good price (if you can still find one - Ebay(?)) but is limited because of that. It may still be adaptable for your project.
I just looked on ebay right now (Not even a thing on voice regognition).· Sorry, but·I don't think that's going to work (The radio shack one). ·I would like it to,·but I don't think it will.· I have been looking around for these chips and my findings were very slim.· The sensory VR stamp looks like the best way to go but it's a 40-pin DIP stamp, and I don't know how I would intergrate it to my BS2 (They show the schematic, but it makes no sence to me).· Also they didn't show the price of it, if any of you guys have a price for it, please show me.· Also, there other ones look good to but take alot of work to get them working.· Wich one is the best one for my project?··
Post Edited (SciTech02) : 4/5/2006 10:48:00 PM GMT
Interesting thought about making the human the LISTENER, and the robot the TALKER, to minimize the amount of listening the robot needs to be able to understand!
Well you don't have to answer the question about how to connect it to my BS2, all I have to do is put the 40-pin stamp on a breadboard (Duh) and have jumper wires go·from my BS2 to the VR Stamp.· I see what you mean· when you say·"get involved", I know you have to·teach it the commands but don't you have to do it with all voice recognition chips (Remember, I want it to do my own commands)?· Oh and thanks·Bates for the link.· Verry clever, but if it was in the middle of a program it would ask those questions every once in a wile, and if you wanted to·tell it to stop you would have to wait for it to ask it.· But it's still clever indead.·
Okay, I found some info on the voice direct 2.· It's perfect for what I'm trying to do!· The site I landed on said the only con was that it was preprogramed, and you would have to pick words from it's unprogramable volcabulary.· Also it said that the text was a little less detailed then they expected, but still detailed.· Overall they said that the problems weren't big enough to affect the rating,·and it gave it a·9.2.
SciTech,
I have a VoiceDirect II board somewhere in my pile that I just haven't had the time to play with, but I do remember that it is a voice-dependent, trainable device (based on my memory of the documentation). You could train it to respond to whatever 16 or 32 commands you like. After training, it responds to commands by bringing a pin or pins high. But, you are correct, they have been discontinued.
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Truly Understand the Fundamentals and the Path will be so much easier...
Oh, I guess I overlooked that.··It's the pits it's·discontinued, but that means they came out with a better one and this one's value will drop, and I could get a good deal on it.· Of corse it could do the exact oposite to...
I should have mentioned in my original post that the Sensory home page has their distributors link on it. Sensory distributors are two: Digi-Key at www.digi-key.com and Semiconductor Store at www.semiconductorstore.com . Hope that helps. Digi-Key lists the Voice Direct II as an obsolete item and to call for availability information. I haven't worked with Semiconductor Store and they don't show the Voice Direct II. Looks like the VR Stamp is the superceeding product. It's worth a call to Digi-Key to see what you can find out though. Let us know.
I gave Digi-Key a call and they listed the VDII as obsolete per Sensory in October of '04, and sold the last of their remaining stock during the fall of '05. Sadly, it is confirmed that they have zero modules on the shelf. Jameco used to sell them too, but no longer list Sensory products in a search there. I'm still looking, but it's looking pretty bleak for the VDII.
The latest 'homework' into the Sensory product line finds an extensive third party development kit available for the VR-Stamp at $129 through a company in the U.K. That development kit which includes a VR-Stamp can be found here:
No problem. I've been following speech recognition and synthisis technologies for many years - it's cool stuff. Actually, I find the 'homework' to be alot of fun and to tell you the truth, I was wondering about the Voice Direct II too. Plus I felt a bit guilty about recomending that you call Digi-Key and then let the rest of us know what the status was. Doesn't that VR-Stamp development board look like a lot of value for the money?
Yes it does but it's still expencive ($130.00 to $50.00).· I like all the features of it but I really don't need that huge developing board, and even if I used it, I would have to·only use it once.· I saw they had another developing board (The VR toolkit), I'll look into that.·
Post Edited (SciTech02) : 4/7/2006 12:55:05 AM GMT
I understand that the expense of the development board isn't practical for a single project. I wanted to throw it out there for nearly the fun factor alone. The VR-Stamp by itself looks like a lot of capability for $40, aside from the development tools. Like I said, I just find the whole topic very interesting.
I see.· I looked for that VR stamp toolkit on the entire web and none are for sale (The only ones are from the Sensory distrbutors and they don't show the price).· I even looked on ebay, they had nothing.··I'll keep looking.··· -SciTech02
Comments
-phar
Yes. I recently saw something on this neat robot called Coyote. It used advanced vision to detect hand gestures, and strong sound systems to hear. It was amazingly advanced. The guys sweep his arm to coyote's right, and says" Coyote; go left". The robot then says in a robot voice:"You pointed one way and told me to go another."-(It was funny to see this reaction) the the guy said"Coyote. Go to my left."The robot amazingly did so
I'm not sure if it is hackable, but hear is a super-ultra mindsttorm bot coming out september that can reconized voice patterns.
http://shop.lego.com/product.asp?p=8527&cn=55&d=11&t=5
Vioce Recognition depends on three things:
Patterns
Amplitude of Sund Waves
Frequency
The best reconizers are the most expensive...
And back on the voice recognition chips, How would you intergrate it to a BS2?· Can you just conect them to the breadboard, or would·you have to completly take it apart and do it?· Then, how would you program it?·
Or better yet, what if I just got a Text-to-speach board and a microphone/speaker, then did some sort of PULSIN command.· It's probably way more complicated then that, but could it be done?··· -SciTech02
http://www.imagesco.com/catalog/hm2007/index.html
http://www.imagesco.com/articles/hm2007/SpeechRecognitionTutorial02.html
http://www.datasheetcatalog.com/datasheets_pdf/H/M/2/0/HM2007.shtml
Thanks.
http://sensoryinc.com/html/company/pr03_03.htm
They also have a new product available called the VR Stamp, and that can be found through the Sensory home page here:
www.sensoryinc.com
Maybe others here have experience with these products and can speak to them directly.
Tim
breaks the sounds down into a phonetic alphabet consisting of:
·
vowels
diphthongs
voiced constants
unvoiced constants
special phonemes
·
The basic idea is to establish frequency matching relationships for each sound in each group and specific pattern matching to a word database.
·
For example the word "Bird" would phonetically look something like· "B-ER-D" where the stored pattern would be key tones within each phoneme.·
·
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Beau Schwabe
IC Layout Engineer
Parallax, Inc.
Pattern matching is not that easy of a field, it was a graduate level course for good reason, the statistics and other mathmatics involved (pattern matching methods) are not that easy to follow without a good background in mathmatics. In the course we wrote a program for mine detection using ground penetrating radar which took the reflected signal, passed it through a 26 bin FFT of radio frequencies, then comapred the 26 dimensional output vector to a sample set of 512 known vectors (256 vector signatures of mined areas and 256 vector signatures of unmined areas). Only about half of our class got better than a 50% (pure chance) guess rate (amusingly one student got it mostly right, but then accidentaly inverted the output boolean, so it was right 11% of the time).
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
6+6=10 (Long live the duodecimal system)
Post Edited (Paul Baker) : 4/4/2006 6:13:30 PM GMT
For example, here is a fairly good write up of Bayes classifier, the first pattern recognition algorithm taught to students studying pattern recognition. But as you see if you don't already understand statistics and some set theory, the write up won't make sense to you.
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
6+6=10 (Long live the duodecimal system)
I suggest that you search for this chip on eBay and other sources of old ICs.
John Wardle
Isle of Man UK
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Manxstamp,
Isle of Man, British Isles
Post Edited (SciTech02) : 4/5/2006 10:48:00 PM GMT
Here is some information on the VCP200 Voice Recognition chip, and one users experience utilizing it. Scan about 1/2 down through the web page:
http://www.seattlerobotics.org/encoder/jul97/gensys.html
Interesting thought about making the human the LISTENER, and the robot the TALKER, to minimize the amount of listening the robot needs to be able to understand!
Regards,
Bruce Bates
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
<!--StartFragment -->
Ryan
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Ryan Clarke
Parallax Tech Support
RClarke@Parallax.com
Post Edited (SciTech02) : 4/6/2006 5:55:24 PM GMT
Tim
I have a VoiceDirect II board somewhere in my pile that I just haven't had the time to play with, but I do remember that it is a voice-dependent, trainable device (based on my memory of the documentation). You could train it to respond to whatever 16 or 32 commands you like. After training, it responds to commands by bringing a pin or pins high. But, you are correct, they have been discontinued.
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Truly Understand the Fundamentals and the Path will be so much easier...
Tim
Tim
http://www.mikroelektronika.co.yu/english/product/tools/easyvrstamp.htm
Looks like you get alot for the money.
Tim
Post Edited (SciTech02) : 4/7/2006 12:43:44 AM GMT
Tim
Post Edited (SciTech02) : 4/7/2006 12:55:05 AM GMT
Tim