Virtual Humans Forum
Virtual Humans Forum
Home | Profile | Register | Active Topics | Members | Search | FAQ
Username:
Password:
Save Password
Forgot your Password?

Virtual Humans
Partner website:
Chatbots.org
 All Forums
 Virtual Humans
 Artificial Intelligence
 Neural Based AI - NBAI (Next Generation AI)
 New Topic  Topic Locked
 Printer Friendly
Previous Page | Next Page
Author Previous Topic Topic Next Topic
Page: of 6

_jc
Curious Member



USA
12 Posts

Posted - Aug 27 2009 :  17:28:24  Show Profile  Visit _jc's Homepage
Hi Grant,
Yes, now that people and their equipment realize the importance of Phase information and the Time Domain, a decent spacial illusion is possible with just stereo speakers (depending on your room acoustics, unless you are very near the speakers).

I wonder if those games you mention use recordings or were totally virtual. I expect that these days good spacial audio is routinely synthesized by software - not something I've investigated though.

Try reversing the polarity of the wires to one of your stereo speakers and listen to the sound field collapse (and you'll probably lose bass). Phase really is important.

Don't know how one neural net could be trained to two tasks. Seems like the last training would mess up the first. I guess you could create any number of single trained neural nets and then sum their outputs. Effect would seem to be the same, right?

----------
_jc
Go to Top of Page

hologenicman
Moderator



USA
3322 Posts

Posted - Aug 27 2009 :  19:02:26  Show Profile  Visit hologenicman's Homepage
Hey there,

quote:
From your above, it sounds like you wish to simulate human hearing. As an expert in electro-acoustics and microphones, I humbly suggest that your concept has a flaw. "Directional mics" (called "cardioid's" in the industry, because of their heart-shaped directional patterns) are very poor microphones technically and I'd avoid them altogether.



I really appreciate this input. Phase is extremely important to me and I will absolutely be axing the directional/cardoid mids from the array.

Yes, I do prefer to set up a human simulation of hearing so I will be persuing your suggestion of the binaural mike set-up when the time comes.

I've started looking at my options, and some small part of me is considering setting up a partition on my hard drive and installing Linux to run my application. The Lazarus/fpc ide/compiler will allow me to port my same generic code over to it and from what I have gathered, no such limitations on the application size exists in Linux.

I am not releasing this application commercially any time soon so I am not tied into Windows. By the time Windows has progressed forward enough in the years to come, I can port my application back to Windows/mac for commercial viabillity.

The other option would be to convert the entire AI script into C++ as a .dll or as a standalone app. I know that Grant would support this, and that C++ probably has a much longer future aheada of it than Pascal probably does a a language, but C++ has not come naturally to me the way that PAscal has.

Still, I know enough C++ to get the job done. Only, I have gone so far down this path with Pascal, that it is a little discouraging to think of hashing it all out again in C++...

Stil, Still..., I really like the pascal write once/compile multiple formats concept. I know that C++ should be protable, and I used WXsidgets for a while as a portable C++ environment, but Pascall seems to be a bit more straight forward.

Mabye, I will have a look at WXwidgets again for a quick diversion.

BTW, don't worry about the "humbly" stuff. We've found around here that we pretty much interact as equals whether they are a teenager, or aged like us or new at this stuff or been fighting with it for decades. As a general rule, most people who follow these ideas are pretty strong willed and we just throw our ideas out there as though we know best. Just put plenty of emoticons in your test if you have any concern about your meaning being missinterpretted.

I absolutely really have benefitted from all of you rsuggestinos thus far and am the better for your input. I'm not about how much I know already, but rather, about how much I can learn from this point on...

Keep me challenged, keep me interested, and don't worry about being humble about it.

John L>
IA|AI

HologenicMan
John A. Latimer
http://www.UniversalHologenics.com

"If the Human brain were so simple that we could understand it,
we would be so simple that we couldn't..."
-Emerson M Pugh-

Current project:http://www.vrconsulting.it/vhf/topic.asp?TOPIC_ID=816&whichpage=1

DISCOVERY: The more I learn, the more I learn how little I know.
GOAL: There's strength in simplicity.
NOTE: Goal not always achieved.

Edited by - hologenicman on Aug 27 2009 19:06:42
Go to Top of Page

GrantNZ
Dedicated Member



New Zealand
2677 Posts

Posted - Aug 28 2009 :  06:18:34  Show Profile
quote:
I wonder if those games you mention use recordings or were totally virtual.

Totally virtual. You can "turn your head" in the game and the audio adjusts to the relative position. Of course it's not as clear as using true surround sound speakers, but the effect is pretty good.

DirectX's sound components have offered 3D simulation for a while, but I've never had the chance to test it out, so I'm not sure if games are using that or some other sound provider.

quote:
Don't know how one neural net could be trained to two tasks. Seems like the last training would mess up the first.

Fair enough.... But then I have to wonder how our own brain manages the task? Or I wonder if a net would/could eventually compartmentalise itself (I suppose with an extreme amount of training) such that it first identifies the type of input and the biases the output towards one "compartment" or another....

quote:
The other option would be to convert the entire AI script into C++ as a .dll or as a standalone app. I know that Grant would support this, and that C++ probably has a much longer future aheada of it than Pascal probably does a a language, but C++ has not come naturally to me the way that PAscal has.

Go with whatever works for you. C++ can be a real pain to learn/use, and in any case porting from one language to another is monkey-work, and at worst can even be farmed out to contractors if/when the commercial time comes. It's more important right now that you're able to reach your goals and feel comfortable with your working environment
Go to Top of Page

hologenicman
Moderator



USA
3322 Posts

Posted - Aug 28 2009 :  08:41:59  Show Profile  Visit hologenicman's Homepage
Thanks, Grant,

I am really more comfortable with the Pascal. I hadn't thought of farming out the conversion work when the time comes. That is a great idea.

I'm going hiking through a 2.5 mile cave with my kids and the Boy Scouts tomorrow, and we are in the middle of our local county fair, but I am looking forward to figuring out the allocating of memory on the heap as soon as I get back on th computer with some free time.

I'm hoping that the allocation process will allow me to keep my current structure and pointers to the arrays so that my throughput processing will be unaffected.

John L>
IA|AI

HologenicMan
John A. Latimer
http://www.UniversalHologenics.com

"If the Human brain were so simple that we could understand it,
we would be so simple that we couldn't..."
-Emerson M Pugh-

Current project:http://www.vrconsulting.it/vhf/topic.asp?TOPIC_ID=816&whichpage=1

DISCOVERY: The more I learn, the more I learn how little I know.
GOAL: There's strength in simplicity.
NOTE: Goal not always achieved.
Go to Top of Page

GrantNZ
Dedicated Member



New Zealand
2677 Posts

Posted - Aug 29 2009 :  07:27:29  Show Profile
Maybe have a look at this page:
http://www.geocities.com/franzglaser/tpsrc/tp_dynarr.html
It includes sections on obtaining memory from the heap, and dynamic arrays (including arrays using heap memory). This might give you a head start.

Enjoy the hike; sounds like fun!!
Go to Top of Page

_jc
Curious Member



USA
12 Posts

Posted - Aug 29 2009 :  17:47:04  Show Profile  Visit _jc's Homepage
@Grant:
On Neural Nets and the human brain. As I understand it the human brain has many specialized areas which communicate with each other. To do something like that with Neural Nets, I would expect something like an analog electronic circuit layout. That is, groups of NNs with their outputs summed, NNs in the feedback loop of other NNs, NNs as filters, combiners, etc., etc.

I guess eventually one would have a master controller program and "sequencer", a place to store intermediate results and such - right, it would be a kind of computer. Have never personally connected one Neural Net to another though.

My interest has shifted somewhat away from NNs to Genetic Algorithms, because of their use in optimizing things. And made no progress because of my other interests and because GAs seem to require programming skills I haven't developed.

Still, NNs and GAs each have their unique place. The great thing about NNs, to my mind, is that they are "black boxes". You need to understand (and correctly format) the input data. And you need samples of correct output data for training. But you don't need to know HOW TO arrive at the answers!

How cool is that? A calculating tool that solves problems that you (or nobody at all) know how to solve? Of course it would be even better if a NN could explain to you how it solved your problem, lol.

----------
_jc
Go to Top of Page

hologenicman
Moderator



USA
3322 Posts

Posted - Aug 30 2009 :  08:41:00  Show Profile  Visit hologenicman's Homepage
Absolutely excellent link, Grant!

quote:
It is a very basic article for beginners only, dealing with type, array and pointer subjects.


This page even ahd a "disclaimer" on it that said that it was only for beginners.

you see, that's the problem. All the other web pages were afraid of puting the information on them because they were afraid of insulting my intelligence. I very much prefer a page that presumes that I don't know anything and teaches it to me from the start.

Again, Thanks.

John L>
IA|AI

HologenicMan
John A. Latimer
http://www.UniversalHologenics.com

"If the Human brain were so simple that we could understand it,
we would be so simple that we couldn't..."
-Emerson M Pugh-

Current project:http://www.vrconsulting.it/vhf/topic.asp?TOPIC_ID=816&whichpage=1

DISCOVERY: The more I learn, the more I learn how little I know.
GOAL: There's strength in simplicity.
NOTE: Goal not always achieved.
Go to Top of Page

Nomeneiste
Intrigued Member



USA
121 Posts

Posted - Sep 22 2009 :  22:11:13  Show Profile  Click to see Nomeneiste's MSN Messenger address
Good morning,

How goes this project, Mr. HologenicMan? Is your system up to snuff, or does it need a boost?

TTFN,
Brandon

- The answer is oorple.
Go to Top of Page

raybe
Curious Member



USA
18 Posts

Posted - May 30 2010 :  05:40:34  Show Profile
JC, does the name Alan Dower Blumlein ring any bells for you. The simplest way of having center is equal left and right. No need for center speaker in true sense of stereo. Surround sound is an effect not any where near actual hearing. Sight and focus plays a large part in actual hearing which leads to the pressure equation of frequencies and time to where the sound is actually coming from. IMO what is behind me acts more like a distraction then unless processed by my brain as something of significance.

raybe
Go to Top of Page

hologenicman
Moderator



USA
3322 Posts

Posted - May 30 2010 :  06:24:06  Show Profile  Visit hologenicman's Homepage
Hi Raybe,

One of the more interesting things about neural nets is their abillity to filter out noise and distractions and draw out the pertinant information.

NN's were used on sonar or radar screens to pick out the objects that were actually significant and remove the random noise. A NN applied over time can make the association of persistance and pick it out from transient noise.

John L>
IA|AI

HologenicMan
John A. Latimer
http://www.UniversalHologenics.com

"If the Human brain were so simple that we could understand it,
we would be so simple that we couldn't..."
-Emerson M Pugh-

Current project:http://www.vrconsulting.it/vhf/topic.asp?TOPIC_ID=816&whichpage=1

DISCOVERY: The more I learn, the more I learn how little I know.
GOAL: There's strength in simplicity.
NOTE: Goal not always achieved.
Go to Top of Page

hologenicman
Moderator



USA
3322 Posts

Posted - May 30 2010 :  06:41:21  Show Profile  Visit hologenicman's Homepage

At first glance, stereo hearing would seem like only two inputs to a neural net (left and right), but each input would be broken out into it's spectrum of pitches with individual intensities.

One could set up 100 individual frequencies per left and right ear by setting 100 Hz steps between 100Hz and 10,000Hz with individual intensity values for each frequency range.

That's 200 auditory neural input values to be fed into the Neural Based AI.

There are other "pre-processing" values that can be extracted as well such as average value for each frequency over time(persistance), peak value for each frequency as well as minimum value for each frequency. That gives 4 values for each frequency range.

There also is a very important biological value of whether the frequency is shifting up or down. This would be extrapulated by comparing the frequency intensity to it's surrounding frequencies over time. That gives two more inupts per frequency(frequency-shift-up and frequency-shift-down).

So, six values per frequency range times 200 frequency ranges totals out at 1,200 audio Neural inputs to be fed into the Neural Based AI.

This could be considered a base resolution for meaningful use of the audio feeds into a Neural Based AI.

John L>
IA|AI

HologenicMan
John A. Latimer
http://www.UniversalHologenics.com

"If the Human brain were so simple that we could understand it,
we would be so simple that we couldn't..."
-Emerson M Pugh-

Current project:http://www.vrconsulting.it/vhf/topic.asp?TOPIC_ID=816&whichpage=1

DISCOVERY: The more I learn, the more I learn how little I know.
GOAL: There's strength in simplicity.
NOTE: Goal not always achieved.
Go to Top of Page

raybe
Curious Member



USA
18 Posts

Posted - May 31 2010 :  00:23:47  Show Profile
Filtering on radar and sonar are also based on time of send and receive. The frequencies change based on the time of send and receive using this info for frequency to determine aspects of the objects it is trying to define making it a little easier to distinguish noise from the frequencies they needed to identify from an object. In acoustics time plays a factor based on length of frequency and the environment that can either diffuse certain frequency or enhance it. The Senheiser setup is another type of analyzer for environments and how frequencies are effected in a certain environments because every person has a different resonance within their body cavities. You can create a desired effect or try to balance the phasing ,cancellations and amplitude of frequencies to a desired effect. Most people that hear flat responses from frequencies in a location, home for example usually dislike that particular response. Many people enjoy a nice base response with some extra high frequency for certain types of music and of coarse this changes with the environment, the music and equipment being used to replicate an overall balance of the sound that would be generalized for most but not all environments according to how it was processed to begin with. It's funny how a company like UREI (I think I spelled it correctly) made a device because of the misuse of equalization by creating the Little Dipper which was a cut only frequency equalizer. Let's face it you just try to get a generalized representation of what we are truly hearing. Wax in your ears can be enough to change any ones perspective of what they hear. Thus studios for recording , mixing and mastering are trying to create either something played by the masses on almost any type of sound system with speakers located in the most general way can still sound enjoyable as well as maintaining quality because of the technology that is affordable to almost any one. Sound is energy effected by so many variables. You wouldn't have enough room in an average house to reproduce a true 60 cycle sound wave but we sure can manipulate enough to go thru our floors and walls and still not hear any high frequency. Bose and Frazier had developed that technique to an art form using special enclosures for their systems. As far as putting cardboard behind your ears I haven't seen or heard of any professional using this technique because again with the use of any multitrack recording and headphones and maybe a plastic type of gobo around the drums because of there dynamic nature and to prevent more bleed thru to all the live mics being used that all of the performances including the ambience of room are mixed again in a control environment. Not to say I haven't made some recordings that sounded just fine without the added process but the idea is to control the frequencies, amplitude and phasing to a more generalized and acceptable quality. I know if I am looking at a screen which I know is front of me and not behind me or I'm not physically a part of the environment, surround sound again just doesn't fool my brain into thinking it is because my brain has already processed all the information it needs to focus on the screen in front of me, no matter how big the screen. Thus the magic of Disney is to submerge you into a fake environment that your brain is fooled into believing what you see and the sounds you hear are originating from different locations thus completing the effect that what you see and hear is believable. The technology that these companies use also work with different types of transducers as close to your ear as possible such as head phones and can still give you either an open room effect or a confined space effect. Headphone technology today can sometimes reproduce these effects better than conventional sound systems that include true environmental effects due to where you are either at home or a noisy night club, small room or large arena. There was a technology used many years ago that would allow people to hear just using there body cavity with no transducers at all but electrical signals that by passed your inner ear. I had papers on it from Neumann many years ago. Very interesting stuff hearing without the inner ear and reproducing sound and effects using your own body.

raybe
Go to Top of Page

hologenicman
Moderator



USA
3322 Posts

Posted - May 31 2010 :  03:42:54  Show Profile  Visit hologenicman's Homepage
Very kool.

I would make a difference between recording methodologies and sound collected for feeding into a Neural Based AI on some level.

quote:
Don't re-invent the wheel. Use the binaural mic, dummy head (having the proper skull resonance) and chest cavity resonance simulator that Sennheiser's experts already developed - or if that's no longer available, something similar. Or, for even more realism, use a human model and put your binaural mic on him or her.

If you want more rear rejection, simply tape cardboard "Mickey Mouse" ears behind the actual outer ears. I've done that while recording performances in a small noisy club and it worked well (except for a few recorded snickers from behind me). That is, the recording without the cardboard "ear extensions" was technically accurate, but not aesthetically so.



The binaural method that JC suggests has great advantages in that it gives mechanical separation of the audio input sources similar to the human experience, however, fidelity is not a real need for this audio feed since a human will not be listening to the sounds of the audio signal.

I suggest that more important than fidelity is "consistency". Whatever system you start feeding into the Neural Based AI needs to stay consistent over time. All the internal associations that are developed will be dependent on the consistency of the audio input system.

Cardboard flaps or super expensive equipment won't make the difference if they are changed around and do not stay consistent.

The NBAI will "experience" it's audio environment, and that experience is what the associations will depend upon, not the clarity of the sound signal.

On the other hand, I imagine that changes in the audio collecting system could be made gradually over time...

John L>
IA|AI


HologenicMan
John A. Latimer
http://www.UniversalHologenics.com

"If the Human brain were so simple that we could understand it,
we would be so simple that we couldn't..."
-Emerson M Pugh-

Current project:http://www.vrconsulting.it/vhf/topic.asp?TOPIC_ID=816&whichpage=1

DISCOVERY: The more I learn, the more I learn how little I know.
GOAL: There's strength in simplicity.
NOTE: Goal not always achieved.
Go to Top of Page

raybe
Curious Member



USA
18 Posts

Posted - May 31 2010 :  04:42:12  Show Profile
I believe you still need more of an electrical consistency rather than a mechanical one in that sense. Any thing mechanical or the bin aural method which just another way of saying stereo but changing the the type of mic that is used will never be consistent. It is the signal after the process that will need to be made to adjust for consistency.

raybe
Go to Top of Page

hologenicman
Moderator



USA
3322 Posts

Posted - May 31 2010 :  05:03:46  Show Profile  Visit hologenicman's Homepage
You may be very right.

Perhaps some form of dynamic compression applied to the electronic signal would help smooth out the inconsistencies from one physical micraphone to the next...

Perhaps, both the physical AND the electrical processing will need to be kept consistent to keep from spasing out the NBAI which will ultimately be processing the signals.

On the other hand, perhaps a solid enough NBAI could obsorb a fair amount of inconsistency and survive. We humans can handle live situations, head phones, echo situations, and a variety of other physically altered circumstances and still process the information.

Although, our ears do still stay attatched to our heads in a relatively similar fashion...

John L>
IA|AI

HologenicMan
John A. Latimer
http://www.UniversalHologenics.com

"If the Human brain were so simple that we could understand it,
we would be so simple that we couldn't..."
-Emerson M Pugh-

Current project:http://www.vrconsulting.it/vhf/topic.asp?TOPIC_ID=816&whichpage=1

DISCOVERY: The more I learn, the more I learn how little I know.
GOAL: There's strength in simplicity.
NOTE: Goal not always achieved.

Edited by - hologenicman on May 31 2010 05:07:41
Go to Top of Page
Page: of 6 Previous Topic Topic Next Topic  
Previous Page | Next Page
 New Topic  Topic Locked
 Printer Friendly
Jump To:
Virtual Humans Forum © V.R.Consulting Go To Top Of Page
This page was generated in 0.47 seconds. Snitz Forums 2000