WAV to MIDI algorithm (Polyphonic).
Michael Davis
I have tried to read posts to answer my question, but I have not
found one. If it exists, please forgive me. I have read may posts,
and maybe a FAQ saying polyphonic WAV to MIDI is impossible. That is
not true.
I have used software such as Intelliscore, but as you probably
know, it does not work well.
I have made an algorithm that can pick the notes out of a
polyphonic WAV file (regardless of instruments involved...although I
have not tried drums), and it is not confused by harmonics.
Unfortunately, its runtime leaves much to be desired. I was wondering
if anyone knows of companies interested in an algorithm such as mine,
or is this pursuit of WAV to MIDI simply an academic exercise? More
time, and hopefully money must be spent to make this algorithm run at
a medium pace.
Regards,
Mike
Laurence Payne
Even if several hours, if it is accurate, you have achieved a great
deal. Are you prepared to have it tested?
> I have made an algorithm that can pick the notes out of a
>polyphonic WAV file (regardless of instruments involved...although I
>have not tried drums), and it is not confused by harmonics.
>Unfortunately, its runtime leaves much to be desired. I was wondering
>if anyone knows of companies interested in an algorithm such as mine,
>or is this pursuit of WAV to MIDI simply an academic exercise? More
>time, and hopefully money must be spent to make this algorithm run at
>a medium pace.
--
My Cubase FAQ page is
www.laurencepayne.co.uk/CubaseFAQ.htm
Feedback welcome.
r_o...@remove_this.hotmail.com
> I have made an algorithm that can pick the notes out of a
>polyphonic WAV file (regardless of instruments involved...although I
>have not tried drums), and it is not confused by harmonics.
Drums do make a mess of things.
What approach did you use to do this?
Robert
( modify address for return email )
Nick White
something
......and in reply I say!:
As Laurence says, it may not matter how fast the thing is, if it
really works. I feel that the biggest lack in programmes that even
come close is proper dealing with expressive parts of the music: Pitch
Bend; vibrato; swell etc.
I will warn you, though that people round here are rather jaded about
all this, and the testing will be tough and thorough. Send it out for
_general_ testing and the results will be rude.
Nick White --- HEAD:Hertz Music
(please remove ns from my header email address to reply)
....damn spam
!!
<")
_/ )
( )
_//- \__/
The trouble with farming is that the flower you stop to smell is probably a weed.
If you _really_ want to live...........farm.
Michael Davis
> So how long does it take to render, say, a minute of music to midi?
> Even if several hours, if it is accurate, you have achieved a great
> deal. Are you prepared to have it tested?
>
Must you embarass me? At its current state, I would say a 500 Mhz
computer can go through one minute of 44.1Khz in about 20 hours or so.
I am working on a distributed component, but that is another story.
Before you mock the algorithm, please remember that I have had very
limited time to optimize it. I can make simple cuts...such as to
22Khz...but my dream is a cut of 99% instead of a puny 50 or so.
Unfortunately, I don't know how much faster it will go anyway. To be
honest, I want it to do even more work so...
For a response to Nick as well as this testing question, I hope you do
not assume too much. The reason I wrote this was because software
such as Intelliscore gave me totally wrong notes. And if the notes
were "right", they were not in the right octave. There is a chance
that some research projects do much better than that, but I did not
see them...so I wrote this. If there are projects exceeding
Intelliscore's recognition that do what mine does, then I've wasted my
time. I can pretty much guarantee that my algorithm can find proper
notes in the octave that your ear would expect them to be in. It
still cannot write a beautiful MIDI score that takes elegant care of
trills, vibrato, etc...and I do think I can solve these problems much
easier than the original problem...but as I said, I am out of time. I
hope I have not given people the wrong impression. I consider this
FAR from being a finished commercial product. Just think about the
runtime! It's hard enough for me to do further work on it when it
takes so long to run. I just want to know if a company is interested
in listening to the theory and developing it seriously.
Regards,
Mike
Robbie Baldock
> I have tried to read posts to answer my question, but I have not
>found one. If it exists, please forgive me. I have read may posts,
>and maybe a FAQ saying polyphonic WAV to MIDI is impossible. That is
>not true.
I've also been looking into signal -> MIDI note value possibilities
recently without much success. However, someone on the Pd mailing
list pointed me in the direction of the fiddle~ object (in the "extra"
folder).
I've had some success in getting it to work (monophonically) but being
a Pd novice my patch is somewhat flaky!
I'm sure you could do something similar using the Java Sound API and a
bit of FFT but I haven't tried this myself yet.
Robbie
Chief Scientist
Aleatory Music Systems
http://www.mp3.com/ams2002/
Michael Davis
approach is new. It has nothing to do with FFTs or anything like
that.
Regards,
Mike
r_obert@REMOVE_THIS.hotmail.com wrote in message news:<64a1kukesgslvj6kf...@4ax.com>...
Bryce Fischer
>Laurence Payne <l...@laurenceDELETETHISpayne.freeserve.co.uk> wrote in message news:<rg11ku0ct0gck2mou...@4ax.com>...
>> So how long does it take to render, say, a minute of music to midi?
>> Even if several hours, if it is accurate, you have achieved a great
>> deal. Are you prepared to have it tested?
>>
>
>Must you embarass me? At its current state, I would say a 500 Mhz
>computer can go through one minute of 44.1Khz in about 20 hours or so.
> I am working on a distributed component, but that is another story.
>Before you mock the algorithm, please remember that I have had very
>limited time to optimize it. I can make simple cuts...such as to
>22Khz...but my dream is a cut of 99% instead of a puny 50 or so.
>Unfortunately, I don't know how much faster it will go anyway. To be
>honest, I want it to do even more work so...
I don't believe Laurence was mocking you. In fact, he was saying that
if what you have accomplished is true, then you have quite an
accomplishment, and should have many suiters to license your
algorythm...
--
Bryce
I don't like having disks crammed into me...
unless they're Oreos, and then only in the mouth
Ian Kemmish
says...
>
>I don't know if I have already covered this, but to my knowledge, my
>approach is new. It has nothing to do with FFTs or anything like
>that.
The first thing you should do is probably try to get advice on IP protection.
A patent is obviously one thing you should investigate if you intend to hawk
your general approach around existing software companies (as opposed to
licensing your code), even if you have ideological reservations about patents.
Confidentical disclosure agreements may not be quite good enough to prevent you
being ripped off.
In any case, the very first thing you should do is write it up, and send it,
registered mail, to yourself (and your solicitor/attorney), but both of you
should keep the mail unopened in a safe place. This will help you to establish
prior art if necessary in the future.
You may also want to investigate some of the companies out there who exist to
help exploit IP. The British Technology Group have a good track record in this
area, but there are surely others.
One way of getting attention might be to write to all the software companies
who are active in this arena, and invite them to send you files that they
think might break your software (carefully explaining the time constraints!).
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
Ian Kemmish 18 Durham Close, Biggleswade, Beds SG18 8HZ, UK
usene...@eeyore.dircon.co.uk Tel: +44 1767 601 361
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
``Behind every successful organisation stands one person who knows the secret
of how to keep the managers away from anything truly important.''
r_o...@remove_this.hotmail.com
It's my belief ( and attempts ) that FFT based approaches cannot do
what he has indicated, except for perhaps one or two notes ... and
even then there are cases that break down.
Regards,
Robert
js
> polyphonic WAV file (regardless of instruments involved...although I
> have not tried drums), and it is not confused by harmonics.
Good stuff!
It's my guess you are using wavelets. Right?
If I was attempting an algorithm for wav to midi, that is the approach I
would take.
Be aware that several patents have been granted relating to wavelet
compression.
I don't know if any have been granted for classification using wavelets.
What ever approach you are using, if you want to reduce the time, you MUST
seek the services of a first class mathematician.
It is always possible to approximate complex methods with much simpler (and
faster) algorithms.
Don't expect much help from a company since most programmers have only
rudimentary mathematical skills.
You will most likely have to contact several universities to find someone
who is not only interested in helping you but has the requisite mathematical
expertise.
good luck.
damian
js
;-)
FA
r_obert@REMOVE_THIS.hotmail.com wrote in message
<9u04kus2okbint6mt...@4ax.com>...
>r...@easynet.co.uk (Robbie Baldock) wrote:
>
>
>It's my belief ( and attempts ) that FFT based approaches cannot do
>what he has indicated, except for perhaps one or two notes ... and
>even then there are cases that break down.
>
Why do you believe that Robert?
If your belief is grounded on frequency resolution issues then there are
ways to mitigate this pretty well using multiresolution techniques like
wavelets and laplacian pyramids.
This works because most instruments are very complex harmonically so even if
you can't resolve the pitch difference between the fundamentals of two
closely spaced notes you are very likely to eventually resolve differences
in harmonics.
In fact you only really have to resort to this sort of thing if you want
"real time" performance (i.e. little delay) if it's not an issue then you
can use a faily large FFT and overlap like billy-o
I'm currently working on a pitch scaling project using such techniques
(multiresolution) and getting pretty good results. To be honest I think that
the biggest two challenges for conversion to MIDI will be detecting the
attack transients polyphonically in order to accurately generate multiple
note on messages and pulling apart the multiple notes (because harmonics
tend to overlap a lot).
With my program I can print out the frequencies of the harmonics and
accurately determine "by eye" the pitches, but it's a pretty big step to get
a dumb machine to do it so I'm concentrating at the moment on my analysis
and rendering spot on for my harmoniser.
I'd guess that a neural network may be an interesting approach cause I think
that detecting multiple notes is likely to be down to subtle interations
with both amplitudes and frequencies that may be hard to isolate
deterministically, but I've no idea of the computational cost of such an
approach.
Cheers, Fraser.
Jeffery S. Jones
Kemmish) wrote:
>In article <a8855dfa.02072...@posting.google.com>, mike...@vt.edu
>says...
>>
>>I don't know if I have already covered this, but to my knowledge, my
>>approach is new. It has nothing to do with FFTs or anything like
>>that.
>
>The first thing you should do is probably try to get advice on IP protection.
>A patent is obviously one thing you should investigate if you intend to hawk
>your general approach around existing software companies (as opposed to
>licensing your code), even if you have ideological reservations about patents.
>Confidentical disclosure agreements may not be quite good enough to prevent you
>being ripped off.
It is is a new enough algorithm, definitely look into patent
protection. While you can copyright the program code, that doesn't
stop someone from using the method in their program if it is unique.
>In any case, the very first thing you should do is write it up, and send it,
>registered mail, to yourself (and your solicitor/attorney), but both of you
>should keep the mail unopened in a safe place. This will help you to establish
>prior art if necessary in the future.
The mailing thing isn't necessary, nor especially useful (there are
too many ways to mess with sealed or unsealed envelopes). An attorney
or other legal official can easily track the date. Copyright
protection of the program itself (in the USA at least) is a trivial
thing by comparison, and also worth doing (make a copy on CDROM,
submit it to the copyright office. The lawyer/solicitor can help with
that too).
Registration makes it easier to prove date of creation, a public
record which, unlike mail, can't be challenged easily.
>You may also want to investigate some of the companies out there who exist to
>help exploit IP. The British Technology Group have a good track record in this
>area, but there are surely others.
>
>
>One way of getting attention might be to write to all the software companies
>who are active in this arena, and invite them to send you files that they
>think might break your software (carefully explaining the time constraints!).
There have been a number of programs (university research projects,
like GAMA for example) which tried to go commercial or shareware with
their conversions. Turning a good idea into a commercially viable
program may not be so easy. The companies already making those sorts
of programs are most likely to be interested in improvements; they
constantly tout their own enhancements.
It isn't so much that Polyphonic is impossible (GAMA has a poly
mode). It is that the results are so much less useful than with
monophonic music; poly off a piano (fairly strongly pitched, only one
note of a given tone at once) is more viable. Even guitar (only six
strings, usually never playing duplicate notes) isn't too bad.
Get up to a full pop song arrangement (and you're not likely to get
one sans drums), and it is impossibly rough. Maybe someday. But even
an orchestral (non-percussion) converter could be useful for improv
recording from poly instruments, small groups, etc.
--
*-__Jeffery Jones__________| *Starfire* |____________________-*
** Muskego WI Access Channel 14/25 <http://www.execpc.com/~jeffsj/mach7/>
*Starfire Design Studio* <http://www.starfiredesign.com/>
*Graphic Reflections and Websites* <http://www.execpc.com/~jeffsj/>
r_o...@remove_this.hotmail.com
>
>r_obert@REMOVE_THIS.hotmail.com wrote in message
><9u04kus2okbint6mt...@4ax.com>...
>>r...@easynet.co.uk (Robbie Baldock) wrote:
>>
>>
>>It's my belief ( and attempts ) that FFT based approaches cannot do
>>what he has indicated, except for perhaps one or two notes ... and
>>even then there are cases that break down.
>>
>
>If your belief is grounded on frequency resolution issues then there are
>ways to mitigate this pretty well using multiresolution techniques like
>wavelets and laplacian pyramids.
>
>This works because most instruments are very complex harmonically so even if
>you can't resolve the pitch difference between the fundamentals of two
>closely spaced notes you are very likely to eventually resolve differences
>in harmonics.
And when the fundamental of one instrument trounces on that of another
... or even the harmonic of the other one ... how to decipher which is
which?
>
>In fact you only really have to resort to this sort of thing if you want
>"real time" performance (i.e. little delay) if it's not an issue then you
>can use a faily large FFT and overlap like billy-o
Will not resolve problem above.
>
>
>I'm currently working on a pitch scaling project using such techniques
>(multiresolution) and getting pretty good results. To be honest I think that
That's the thing ... "pretty good results". There are a variety of
frequency based programs that will work under certain conditions, but
fail under many others.
>the biggest two challenges for conversion to MIDI will be detecting the
>attack transients polyphonically in order to accurately generate multiple
>note on messages and pulling apart the multiple notes (because harmonics
>tend to overlap a lot).
>
>With my program I can print out the frequencies of the harmonics and
>accurately determine "by eye" the pitches, but it's a pretty big step to get
>a dumb machine to do it so I'm concentrating at the moment on my analysis
>and rendering spot on for my harmoniser.
Very interesting work, even if it doesn't pan out.
>
>I'd guess that a neural network may be an interesting approach cause I think
>that detecting multiple notes is likely to be down to subtle interations
>with both amplitudes and frequencies that may be hard to isolate
>deterministically, but I've no idea of the computational cost of such an
>approach.
So, you've now left the FFT domain, which was my original point.
Regards,
Robert
Michael Davis
Thank you for all of your information. The more protection I can
find, the better...as I doubt anyone would trust a few test results
without knowing the details of its operation. Still, I must admit
that patents concern me for one main reason. What if someone reads
your patent, makes their algorithm act like yours, but refuses to tell
anyone how theirs works? Couldn't they sell your algorithm in secret?
To be honest, I agree with Jeffrey that making anything into a
commercial product is quite difficult.
For the record, I am very pleased to see that there are others with
such an interest in making things work. When people say things
"can't" and "can never be done", it bothers me.
Michael Davis
No, I am not using wavelets. I don't even think it's related, but
then again, I'm not an expert on wavelets. As far as your comments
about a mathematician, I fear you are correct. I should probably
spend some time finding one. I also agree with another post I saw
about the FFT having problems with differentiating harmonics. The
more "tweaking" one does a technique, the more it becomes questionable
to me. Because of the speed, I am not sure mine will be useful on its
own compared to other techniques that run so fast...but its strong
point is dealing with harmonics and various random instruments.
Determining if something is a harmonic is very straightforward.
This hush-hush stuff does bother me to a certain degree, I must
admit. Being a scientist (or whatever is below that), it pains me to
keep secrets. If the world were not about money, many more things
would be made.
I keep thinking about drums lately too because of the comments about
them. I would like to run one if anyone is interested. Then again, I
have no drum. Is a bell the same as a drum? If so, is a pint glass
the same as a bell? I do have one of those...
Regards,
Mike
robert bristow-johnson
>
> > I have made an algorithm that can pick the notes out of a
> > polyphonic WAV file (regardless of instruments involved...although I
> > have not tried drums), and it is not confused by harmonics.
>
> Good stuff!
> would take.
> Be aware that several patents have been granted relating to wavelet
> compression.
> I don't know if any have been granted for classification using wavelets.
>
> What ever approach you are using, if you want to reduce the time, you MUST
> seek the services of a first class mathematician.
> It is always possible to approximate complex methods with much simpler (and
> faster) algorithms.
> Don't expect much help from a company since most programmers have only
> rudimentary mathematical skills.
> You will most likely have to contact several universities to find someone
> who is not only interested in helping you but has the requisite mathematical
> expertise.
damian, you are quite the comedian, ain't ya?
--
r b-j
r...@surfglobal.net a.k.a. rob...@wavemechanics.com
"Don't give in to the Dark Side. Boycott intel and microsoft."
r_o...@remove_this.hotmail.com
>Determining if something is a harmonic is very straightforward.
Do you wish to just make that statement, and then stop? If yes,
perhaps it is better to not to make it, if you are not prepared to
share your reasoning and followup logic.
Somethings are proprietary, yes. Others are shared on technical
groups like these for the mutual benefit of all ... and in return for
previous knowledge gained.
Robert
Jeffery S. Jones
>Ian and Jeffrey,
> Thank you for all of your information. The more protection I can
>find, the better...as I doubt anyone would trust a few test results
>without knowing the details of its operation. Still, I must admit
>that patents concern me for one main reason. What if someone reads
>your patent, makes their algorithm act like yours, but refuses to tell
>anyone how theirs works? Couldn't they sell your algorithm in secret?
Until they get caught, then patent infringement lawsuits can make it
quite painful. The advantage with the patent is the method is
published, and legitimate software makers will be happy to deal with
licensing it if it works.
OTOH, patents are harder to get than copyrights. Any computer
program can be copyrighted. Patent requires you to go through an
involved process to prove that the algorithm is new and unique (not an
extrapolation of existing patents), and is sufficiently innovative to
qualify as a new machine/technology. Patent protection is also only a
few years (varies on country and type of patent).
> To be honest, I agree with Jeffrey that making anything into a
>commercial product is quite difficult.
> For the record, I am very pleased to see that there are others with
>such an interest in making things work. When people say things
>"can't" and "can never be done", it bothers me.
It is hard to do, and difficult to do effectively. Many programs
have come out with claims which exceeded typical results. Those which
make more modest claims are more acceptable, but then how do they talk
people into paying good money for a program with limited use? ;-)
It isn't that programs like Intelliscore aren't useful. It is just
that the situations where it is effective seem about the same as the
few hardware audio pitch to MIDI devices.
Netmask
question or maybe not asking the question? A concrete example:
The didgeridoo does not exist in the standard midi protocol, there are many
instruments that don't exist in the midi protocol so what happens to a wav
file recorded in the Taj (with its 8 seconds or so reverb) of 2 didgeridoos
when you "convert' it to midi? Non western tonality, countless reflections
and no corresponding instrument? So I am prepared to say no one will ever
faithfully and accurately convert a complex wav file into a midi file -
never! Someone will eventually make a program that does a reasonable
translation that may be a guide for transcription purposes. Look out folks
here comes the spooks!!! CIA. MI6 etc Eshalon etc. You have a recording of a
top secret conversation and you convert it to a midi file - Henry Kissinger
aka Steinway - slip the midi file to a fellow spook who recovers Henry from
the bass clef? nah! wasted time! silly stuff. It bothers me when I see
people on a quest for some holy thingy who want to believe everything is
possible! Tilting at Windmills. The question remains WHY?
"Michael Davis" <mike...@vt.edu> wrote in message
news:a8855dfa.02072...@posting.google.com...
Andreas Tell
there are some other methods for testing the presence of a harmonic
structure with given pitch. One of my favourites in means of simplicity is a
multiband autocorrelation method. It is a quite time-consuming technique but
can be optimized.
=================================================
Test presence of harmonic structure based at w_0 :
1) Harmonic Filtering
filter the signal with abs(H(exp(iw))=(1/2+cos(2*pi*w/w_0)/2)^N
where N>=1 is a fine-tune-parameter.
2) Harmonic Enhancing
Apply a nonlinear function to the resulting signal. I had good results with
a simple square-function
3) Autocorrelation
Do a short-time-autocorrelation and test for a maximum at exactly t=2pi/w_0.
If it's there, check if it stronger than the autocorrelation value at
t=4pi/w_0 and t=pi/w_0. If this is true, your test is positive.
=================================================
I found this method during my research on polyphonic pitch detection. I
always wanted to write a paper about it, but did not have the time to do so.
You are free to use it for anything you would like, but please tell me if
you do so.
All the best
Andreas Tell
theoretical Physics
University of Konstanz
Nick White
wrote something
......and in reply I say!:
>mike...@vt.edu (Michael Davis) wrote:
>
>>Determining if something is a harmonic is very straightforward.
>
>Do you wish to just make that statement, and then stop? If yes,
>perhaps it is better to not to make it, if you are not prepared to
>share your reasoning and followup logic.
>
>Somethings are proprietary, yes. Others are shared on technical
>groups like these for the mutual benefit of all ... and in return for
>previous knowledge gained.
...and you have shared what and where? Are you asking this guy to
start revealing his methods?
robert bristow-johnson
jef...@execpc.com wrote on 07/28/2002 00:47:
> OTOH, patents are harder to get than copyrights. Any computer
> program can be copyrighted. Patent requires you to go through an
> involved process to prove that the algorithm is new and unique (not an
> extrapolation of existing patents), and is sufficiently innovative to
> qualify as a new machine/technology.
i wish. there are certainly far more patents that fail that criterion
("new" and "unique" and not an extrapolation of existing patents) than those
that satisfy it. i think that patent examiners check to see that the patent
"looks okay" in that the claims are made in the correct dialect of legalese
(they might say unambiguous language) but i don't think that the patent
examiners nix a patent application based on the content or the novelty or
feasibility unless you put the words "perpetual motion machine" in it
somewhere.
> Patent protection is also only a
> few years (varies on country and type of patent).
it used to be 17 (years) in the U.S. but i have heard recently it's gone up
to 20. that's a pretty long time.
--
r b-j
Wave Mechanics, Inc.
45 Kilburn St.
Burlington VT 05401-4750
tel: 802/951-9700 ext. 207 http://www.wavemechanics.com/
fax: 802/951-9799 rob...@wavemechanics.com
--
r_o...@remove_this.hotmail.com
>On Sat, 27 Jul 2002 21:28:53 -0700, r_obert@REMOVE_THIS.hotmail.com
>wrote something
>......and in reply I say!:
>
>>mike...@vt.edu (Michael Davis) wrote:
>>
>>>Determining if something is a harmonic is very straightforward.
>>
>>Do you wish to just make that statement, and then stop? If yes,
>>perhaps it is better to not to make it, if you are not prepared to
>>share your reasoning and followup logic.
>>
>>Somethings are proprietary, yes. Others are shared on technical
>>groups like these for the mutual benefit of all ... and in return for
>>previous knowledge gained.
>
>...and you have shared what and where? Are you asking this guy to
>start revealing his methods?
Shared plenty on this group Nicky ... for several years now. Asked
him not to boast, if he wasn't gonna back it up ... not disclose
proprietary methods ... big difference pal. But I think that was
pretty apparent from my post.
Anyway, we chatted via email, and cleared it up. Thanks for your
concern though.
Cheers,
Robert
Nick White
something
......and in reply I say!:
OK. I can see the need for hush-hush. Even though you _may_ not be
driven by money, it would piss anyone off to see some other person rip
you off.
What you need to do is _repeatedly_ provide results, from complex
waves to MIDI. Do you have a website? Post MIDIs of well-known stuff
there. If you posted only short exerpts, then copyright would be
_allowable_. If you posted MIDI so classical pieces that are no longer
copyrighted then you should be safe. Someone correct me hwere if I am
wrong.
While the slowness of the algorithm is of some consideration,
primarily, you need to be _completely_ honest about how much time
_you_ put into each piece of transcription, as distinct from the time
the programme takes. In other words, the result should be exactly as
the programme makes it, not tweaked by you to make it sound better.
Believe me, your honesty will maybe make things tough in the short
term, but if you produce false impressions now you _will_ pay in the
from of lambasting later.
Then the next step would be to approach people, whom you think may be
able to make this a going concern, with _results_. I feel that you
still do not need to produce methods. At this stage I woudl have
though _somebody_ would be willing to take it on on the basis that
_if_ you can do this _then_ such and such a bargain can be struck.
Nick White
<UseAddr...@pandora.orbl.org> wrote something
>On Mon, 29 Jul 2002 04:57:52 GMT, in
><3d448908....@news.m.iinet.net.au>, nsnf...@iinet.net.au
>(Nick White) wrote:
>
>>On Sat, 27 Jul 2002 21:28:53 -0700, r_obert@REMOVE_THIS.hotmail.com
>>wrote something
>>......and in reply I say!:
>>
>>>mike...@vt.edu (Michael Davis) wrote:
>>>
>>>>Determining if something is a harmonic is very straightforward.
>>>
>>>Do you wish to just make that statement, and then stop? If yes,
>>>perhaps it is better to not to make it, if you are not prepared to
>>>share your reasoning and followup logic.
>>>
>>>Somethings are proprietary, yes. Others are shared on technical
>>>groups like these for the mutual benefit of all ... and in return for
>>>previous knowledge gained.
>>
>>...and you have shared what and where? Are you asking this guy to
>>start revealing his methods?
>
>I don't think it's a pure matter of asking the guy to reveal his
>methods as much as it is a request to back his assertions.
>Otherwise continuing this thread really is rather pointless and
>our time could be better spent reading the old story, "The
>Emperor's New Clothes."
You're probably right.
Nick White
wrote something
......and in reply I say!:
OK....Bobby. Whatever...
I get the impression from your reply that I am not that wrong about
the tone of your original message.
r_o...@remove_this.hotmail.com
>On Mon, 29 Jul 2002 23:54:58 GMT, r_obert@REMOVE_THIS.hotmail.com
>wrote something
>......and in reply I say!:
>
>OK....Bobby. Whatever...
>
>I get the impression from your reply that I am not that wrong about
>the tone of your original message.
Wrong again, bucko. Lol.
Netmask
years ago called "Why?" can't recall the author but was to the point..
"Jim Higgins" <UseAddr...@pandora.orbl.org> wrote in message
news:kllakus1jfvcl3r2q...@4ax.com...
> On Mon, 29 Jul 2002 04:57:52 GMT, in
> <3d448908....@news.m.iinet.net.au>, nsnf...@iinet.net.au
> (Nick White) wrote:
>
> >On Sat, 27 Jul 2002 21:28:53 -0700, r_obert@REMOVE_THIS.hotmail.com
> >wrote something
> >......and in reply I say!:
> >
> >>mike...@vt.edu (Michael Davis) wrote:
> >>
> >>>Determining if something is a harmonic is very straightforward.
> >>
> >>Do you wish to just make that statement, and then stop? If yes,
> >>perhaps it is better to not to make it, if you are not prepared to
> >>share your reasoning and followup logic.
> >>
> >>Somethings are proprietary, yes. Others are shared on technical
> >>groups like these for the mutual benefit of all ... and in return for
> >>previous knowledge gained.
> >
> >...and you have shared what and where? Are you asking this guy to
> >start revealing his methods?
>
> I don't think it's a pure matter of asking the guy to reveal his
> methods as much as it is a request to back his assertions.
> Otherwise continuing this thread really is rather pointless and
> our time could be better spent reading the old story, "The
> Emperor's New Clothes."
>
> --
> Jim Higgins, quasimodo AT yahoo DOT com
>
> alt.music.midi FAQ - http://home.sc.rr.com/cosmogony/ammfaq.html
robert bristow-johnson
r_obert@REMOVE_THIS.hotmail.com at r_obert@REMOVE_THIS.hotmail.com wrote on
07/29/2002 19:54:
...
>
> Shared plenty on this group Nicky ... for several years now.
hi R_obert,
if you haven't, check the "newsgroup" header. this is cross-posted to a
couple of other ngs which are almost certainly more noisy than comp.dsp.
Nick has no idea (unless he's sophisticated enough to check you out on
Google groups) what you've been doing on comp.dsp.
regrads,
r b-j
Martin
Davis <mike...@vt.edu> writes
> This hush-hush stuff does bother me to a certain degree, I must
>admit. Being a scientist (or whatever is below that), it pains me to
>keep secrets. If the world were not about money, many more things
>would be made.
Then don't worry about it.
A full patent will take about 2 years and approx. a minimum of $100k
in legal and search fees. If you feel you want to go that route then do
so, but it'll bring you more sleepless nights than you could ever
imagine, and you'll spend more time dealing with the patent than on the
product.
OTOH, start publishing your research (or at least putting it into some
kind of order) and you start to build a reputation, and that will
probably be worth more in term of self-esteem and income than any
patent.
--
Martin of Strawberry Hill
Randy Yates
I agree. Unfortunately, most of us work for companies who don't.
--
Randy Yates
DSP Engineer, Sony Ericsson Mobile Communications
Research Triangle Park, NC, USA
randy...@sonyericsson.com, 919-472-1124
robert bristow-johnson
martin...@paladio.demon.co.uk wrote on 08/04/2002 13:05:
> In message <a8855dfa.02072...@posting.google.com>, Michael
> Davis <mike...@vt.edu> writes
>> This hush-hush stuff does bother me to a certain degree, I must
>> admit. Being a scientist (or whatever is below that), it pains me to
>> keep secrets. If the world were not about money, many more things
>> would be made.
>
> Then don't worry about it.
> A full patent will take about 2 years and approx. a minimum of $100k
> in legal and search fees.
that sound excessively high. if the claims of a patent go unchallenged, we
were quoted (by a pretty reputable patent attorney in Boston) US$22K. that
was the minimum. even if there are problems, i would be surprized to see it
go as high as $100K.
> If you feel you want to go that route then do
> so, but it'll bring you more sleepless nights than you could ever
> imagine, and you'll spend more time dealing with the patent than on the
> product.
> OTOH, start publishing your research (or at least putting it into some
> kind of order) and you start to build a reputation, and that will
> probably be worth more in term of self-esteem and income than any
> patent.
the other thing that publishing it does is it establishes it as prior art so
that some other jerk cannot patent your idea and prevent *you* from using it
(or charging you royalties to use your own idea). if you protect your idea
by use of "trade secret", then you better make and market some product that
uses it so, if you ever had to, you could use that to establish prior art.
also, if the idea is *not* likely to be a big breakthrough and super
money-maker (at least not enough of a money maker to justify the outa pocket
cost of patenting), publishing might be a good thing to do.
those were some of the reasons we did the IEEE Mohonk paper on frequency and
sweep-rate estimations of time-variant sinusoids last year. if it made a
*big* difference in the sound of a time-scaling phase vocoder (from the
prior state of the art), there is no way my boss would let me publish it or
tell anyone about it. so it turned out to be more mathematically useful
than practically. oh well. at least no one else can patent it.