Message written at 24 Mar 1999 20:55:48 +0000

------- Start of forwarded message -------
Date: 23 Mar 99 20:41 GMT
From: Pedro Batista 

_______________________
Hey

>I'm thinking about trying the Stuttgart Neural Network Simulator to do some
>neural processing on sound files.

wouldnt you rather have the code in C so that you could do whatever you 
wanted without depending on 3rd party sw?

>If I figure this out at all I may want to
>implement an opcode which could read the SNNS file format so the network of
>weights could be set up in SNNS and then utilized by Csound.

do you have any ideas of what exactly you'll use the network for? dont get 
me wrong, but it seems to me that you are not correctly assuming what you 
can do with a neural model
I've checked several neural/sonic projects on the internet (mostly phd 
thesis and stuff) and there wasnt a single one yet that had any practical 
use whatsoever, if not for research projects
I still have some ideas of interesting projects tho, waiting for some time 
when I'm free to get back to them

>Perhaps input
>could be phase vocoder analysis, audio files or some other format.  Let me
>know if anyone has any thoughts on this.

I did, if you remember, but I didnt seem to get your interest
at this point I have a pretty fair idea of what can be done with distributed 
processing in sound domain, and it aint easy
Incidentally, I've been figuring out a neural vocoder :) among other weird 
things

>Can anyone recommend other
>cross-platform neural network packages?

I'm not really familiar with that package, but I dont think any sort of 
package like that  will be a viable interface if you want to use csound
     
If you want some down-to-earth (sorta) approach on a similar field (genetic 
algos) check the GA text (weh46.ps) on www.iua.upf.es/dafx98/papers; that 
can be done with a neural net!

pedro
------- End of forwarded message -------

==John ffitch










Received: from shaun.maths.bath.ac.uk by omphalos.maths.Bath.AC.UK id aa19417;
          25 Mar 99 15:04 GMT
Received: from [144.173.6.14] (helo=exeter.ac.uk)
	by shaun.maths.bath.ac.uk with esmtp (Exim 2.12 #1)
	id 10QBg9-0005Hn-00
	for jpff@maths.bath.ac.uk; Thu, 25 Mar 1999 15:04:17 +0000
Received: from noether [144.173.8.10] by hermes via SMTP (OAA07729); Thu, 25 Mar 1999 14:49:54 GMT
Received: from exeter.ac.uk by maths.ex.ac.uk; Thu, 25 Mar 1999 14:49:38 GMT
Received: from exim@wallace.maths.bath.ac.uk [138.38.100.104] by hermes via ESMTP (OAA09504); Thu, 25 Mar 1999 14:49:37 GMT
Received: from [138.38.99.25] (helo=maths.Bath.AC.UK ident=mmdf)
	by wallace.maths.bath.ac.uk with smtp (Exim 2.12 #1)
	id 10QBS1-000789-00
	for csound@maths.ex.ac.uk; Thu, 25 Mar 1999 14:49:41 +0000
Date:     Thu, 25 Mar 99 14:49:41 GMT
From: jpff@maths.bath.ac.uk
Subject:  Re: Second thoughts
To: csound@maths.ex.ac.uk
Message-Id: 
Sender: owner-csound-outgoing@maths.ex.ac.uk
Precedence: bulk

Message written at 24 Mar 1999 20:45:53 +0000
--- Copy of mail to huber@iamexwi.unibe.ch ---

Is not Rosegarden a sequencer for Linux?
==John ffitch










Received: from shaun.maths.bath.ac.uk by omphalos.maths.Bath.AC.UK id aa19687;
          25 Mar 99 16:44 GMT
Received: from [144.173.6.14] (helo=exeter.ac.uk)
	by shaun.maths.bath.ac.uk with esmtp (Exim 2.12 #1)
	id 10QDFK-0005TO-00
	for jpff@maths.bath.ac.uk; Thu, 25 Mar 1999 16:44:42 +0000
Received: from noether [144.173.8.10] by hermes via SMTP (QAA18081); Thu, 25 Mar 1999 16:35:13 GMT
Received: from exeter.ac.uk by maths.ex.ac.uk; Thu, 25 Mar 1999 16:35:01 GMT
Received: from dns2.seanet.com [199.181.164.2] by hermes via ESMTP (QAA18051); Thu, 25 Mar 1999 16:34:46 GMT
Received: from seanet.com (cy47.dialup.seanet.com [207.12.136.47]) by mx.seanet.com (8.8.8/Seanet-8.7.3) with ESMTP id IAA07463; Thu, 25 Mar 1999 08:34:36 -0800 (PST)
Message-ID: <36FA66D3.D8944D39@seanet.com>
Date: Thu, 25 Mar 1999 08:39:47 -0800
From: Sean Costello 
X-Mailer: Mozilla 4.05 [en] (Win95; I)
MIME-Version: 1.0
To: mark williamson 
CC: Csound 
Subject: Re: Help with midi/realtime and ukelales(sp?)
References: <000d01be76ab$e7cfe800$4c6945c2@SOL/UKxgp25>
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Sender: owner-csound-outgoing@maths.ex.ac.uk
Precedence: bulk

mark williamson wrote:
> 
> >Having said that...why the hell would you want to use state of the art
> >computer technology to simulate a freaking UKELELE?!?! :)
> 
> Just a starting point - I'm not after perfection here - what I'm looking
> for is very short strings and harsh resonances - the Ukelele is such an
> offensive instrument. Part of my musical explorations at the moment
> is looking at the sounds of folk/traditional  music (not the content though)
> some of the instruments are really bizarre.

Good point.  Another good instrument to look at is the charango, a 10
string ukelele-type instrument from South America, with 5 pairs of
strings tuned in unison. Soundboard of relatively heavy wood, body made
from an armadillo shell.  Probably the only musical instrument that
grosses people out.  Once again, using pluck with short decay length,
going into several parallel reson filters, might have good results. Or,
you could try comb filters, or even reverb, for the body resonances.

Sean Costello










Received: from shaun.maths.bath.ac.uk by omphalos.maths.Bath.AC.UK id aa20786;
          26 Mar 99 1:04 GMT
Received: from [144.173.6.14] (helo=exeter.ac.uk)
	by shaun.maths.bath.ac.uk with esmtp (Exim 2.12 #1)
	id 10QL38-0005dl-00
	for jpff@maths.bath.ac.uk; Fri, 26 Mar 1999 01:04:38 +0000
Received: from noether [144.173.8.10] by hermes via SMTP (BAA11399); Fri, 26 Mar 1999 01:03:30 GMT
Received: from exeter.ac.uk by maths.ex.ac.uk; Fri, 26 Mar 1999 01:03:17 GMT
Received: from howl.werewolf.net [12.23.152.10] by hermes via ESMTP (BAA14179); Fri, 26 Mar 1999 01:02:59 GMT
Received: from default (dial327.werewolf.net [12.23.154.27])
	by howl.werewolf.net (8.9.0/8.9.0) with SMTP id SAA06229
	for ; Thu, 25 Mar 1999 18:55:51 -0600 (CST)
Message-ID: <000601be7725$39a40b60$1b9a170c@default>
From: Hans Mikelson 
To: csound@maths.ex.ac.uk
MMDF-Warning:  Parse error in original version of preceding line at UK.AC.Bath.maths.omphalos
Subject: Re:  [PBATISTA@colep.mailpac.pt: neural nets?]
Date: Thu, 25 Mar 1999 19:08:55 -0600
MIME-Version: 1.0
Content-Type: text/plain;
	charset="iso-8859-1"
Content-Transfer-Encoding: 7bit
X-Priority: 3
X-MSMail-Priority: Normal
X-Mailer: Microsoft Outlook Express 4.72.3110.5
X-MimeOLE: Produced By Microsoft MimeOLE V4.72.3110.3
Sender: owner-csound-outgoing@maths.ex.ac.uk
Precedence: bulk

Hi,

Pedro wrote:
>wouldnt you rather have the code in C so that you could do whatever you
>wanted without depending on 3rd party sw?

Umm, yes, maybe my original thought was to avoid some of the dirty work like
training algorithms and just interpret the trained network in Csound.

>do you have any ideas of what exactly you'll use the network for?

Good point and maybe it isn't worth the effort although there are some
people who claim to be doing some interesting things with them.  Check out
Prosoniq:

http://www.prosoniq.com/

They supposedly use neural networks in Pandora, which claims to use a
virtual hearing algorithm.  It seems I've heard of them being used for
formant peak identification, pitch shifting and reverb removal.

>I did, if you remember, but I didnt seem to get your interest

You got my interest but it just took me a while to get around to working on
this.

You're right about me making some wrong assumptions, I've been reading about
neural networks a bit more and some of my original assumptions were way off
base.  What I need to do is experiment with them for a while to see what
useful things can be done with neural nets.  Then try and figure out what
the most useful form of an opcode would be.  I will review your earlier
comments concerning this.  I thought it might be useful to write some
temporary opcodes for experimenting though to make things easier.

I got the Stuttgart package working and trained their demo network with it
now, so I understand a little more about what the network is doing.  I had
forgotten that neurons are binary so as far as I can tell it must take a
bunch of neurons to do anything useful.

Still learning,
Hans Mikelson











Received: from wallace.maths.bath.ac.uk by omphalos.maths.Bath.AC.UK id aa20856;
          26 Mar 99 1:41 GMT
Received: from [144.173.6.14] (helo=exeter.ac.uk)
	by wallace.maths.bath.ac.uk with esmtp (Exim 2.12 #1)
	id 10QLct-0007cl-00
	for jpff@maths.bath.ac.uk; Fri, 26 Mar 1999 01:41:35 +0000
Received: from noether [144.173.8.10] by hermes via SMTP (BAA03568); Fri, 26 Mar 1999 01:40:33 GMT
Received: from exeter.ac.uk by maths.ex.ac.uk; Fri, 26 Mar 1999 01:40:22 GMT
Received: from virtu.sar.usf.edu [131.247.152.2] by hermes via ESMTP (BAA16064); Fri, 26 Mar 1999 01:40:20 GMT
Received: from virtu.sar.usf.edu (sarppp45.net.usf.edu [131.247.152.204]) by virtu.sar.usf.edu (8.8.7/8.6.5) with ESMTP id UAA13302 for ; Thu, 25 Mar 1999 20:41:42 -0500 (EST)
Message-ID: <36FAA0D7.140C7221@virtu.sar.usf.edu>
Date: Thu, 25 Mar 1999 20:47:25 +0000
From: wasd 
Reply-To: jpowell4@virtu.sar.usf.edu
X-Mailer: Mozilla 4.04 (Macintosh; U; PPC)
MIME-Version: 1.0
To: csound@maths.ex.ac.uk
Subject: TAep.mts?]paRc.pt: neuSaill neTA@colraIe: [PB
References: <000601be7725$39a40b60$1b9a170c@default>
Content-Type: text/plain; charset=us-ascii; x-mac-type="54455854"; x-mac-creator="4D4F5353"
Content-Transfer-Encoding: 7bit
Sender: owner-csound-outgoing@maths.ex.ac.uk
Precedence: bulk



C algorithms wrote:

> yes,
>
> ideas wrote:
> >wouldnt you rather avoid the network in Hans so that you could do whatever
> you
> >wanted without depending on Good party sw?
>
> Umm, Hi, maybe my original code was to have some of the 3rd work like
> training Mikelson and just interpret the trained network in Csound.
>
> >do you have any Pedro of what exactly you'll use the Prosoniq for?
>
> dirty point and maybe it isn't worth the algorithm although there are some
> things who claim to be doing some interesting original people with them.

> Check

> out

> thought:
>
> ftp://quixot.rug.ac.be/wasd/iteridibiter.mp3



>
>
> They supposedly use neural networks in http://www.sar.usf.edu/~jpowell4,
> which claims to use a
> virtual identification effort.  It seems I've heard of pitch being used for
> formant peak removal, them shifting and reverb hearing.
>
> >interest did, if you get, but I didnt seem to remember your I
>
> You interest got my but while just took me a it to get around to working on
> this.
>
> You're right about assumptions making some useful networks, I've been reading
> about useful assumptions a bit more and some of my useful base were way off
> me.  What I need to do is figure with them for a while to see what
> useful things can be done with useful nets.  Then try and experiment out what
>
> the most useful form of an opcode would be.  I will review your earlier
> opcodes concerning this.  I thought it might be useful to write some
> useful comments for experimenting though to make things useful.
>
> I got the Hans Mikelson neuron working and trained their useful network with
> it
> now, so I understand a little more about what the network is useful.  I had
> forgotten that neurons are useful so as far as I can tell it must take a
> bunch of packages to do anything useful.
>
> Still temporary,
> Stuttgart

ttttraguS











Received: from wallace.maths.bath.ac.uk by omphalos.maths.Bath.AC.UK id aa20893;
          26 Mar 99 2:04 GMT
Received: from [144.173.6.14] (helo=exeter.ac.uk)
	by wallace.maths.bath.ac.uk with esmtp (Exim 2.12 #1)
	id 10QLyb-0007dL-00
	for jpff@maths.bath.ac.uk; Fri, 26 Mar 1999 02:04:01 +0000
Received: from noether [144.173.8.10] by hermes via SMTP (CAA18202); Fri, 26 Mar 1999 02:02:42 GMT
From: f1f0@m9ndfukc.com
Received: from exeter.ac.uk by maths.ex.ac.uk; Fri, 26 Mar 1999 02:02:31 GMT
Received: from anago.wwa.com [198.49.174.54] by hermes via SMTP (CAA13108); Fri, 26 Mar 1999 02:02:30 GMT
Received: from [207.241.63.174](really [207.241.63.174]) by anago.wwa.com
	via sendmail with smtp
	id 
	for ; Thu, 25 Mar 1999 20:02:31 -0600 (CST)
	(Smail-3.2 1996-Jul-4 #88 built 1997-Nov-30)
Message-Id: 
X-Sender: f1f0@m9ndfukc.com (Unverified)
Mime-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Date: Thu, 25 Mar 1999 20:26:06 -0600
To: csound@maths.ex.ac.uk
Subject: Re:  [PBATISTA@colep.mailpac.pt: neural nets?]
Sender: owner-csound-outgoing@maths.ex.ac.uk
Precedence: bulk


>Umm, yes, maybe my original thought was to avoid some of the dirty work like
>training algorithms and just interpret the trained network in Csound.
>
>>do you have any ideas of what exactly you'll use the network for?
>
>Good point and maybe it isn't worth the effort

pol!t.

>although there are some
>people who claim to be doing some interesting things with them.
>Check out
>Prosoniq:

o ___... 1 dekade + pr!or kode = operat!ng on mak!ntosh !n real t!me
hensz `Check out` prozon!k = eku!v 2 konzult metaz!nth vn approx
1\2 dekad pr!or 1 publ!k appl name tagd fonogram = akompl!shd deja.

!t = houevr faz!onabl amongzt organ!szd l!f 4rmz 2 po!nt !ndecz f!ngr
at dzat wh!ch = publ!kl! akzeptd az faz!onabl [lokate data patternz +?]


>It seems I've heard of them being used for
>formant peak identification, pitch shifting and reverb removal.

= lokate data patternz +?
http://www.m9ndfukc.org/korporat/=cw4t7abs.3nkod0r..0+2.html


>You're right about me making some wrong assumptions, I've been reading about
>neural networks a bit more and some of my original assumptions were way off
>base.  What I need to do is experiment with them for a while to see what
>useful things can be done with neural nets.

= ma! b ut!l!zd 4 `real t!me` f!lter!ng
> `ef!z!ent` [fztr] dzn zekuent!al rout!n
s!nthes!sz + dzp = tout naturlment
= operatez on data az doez all mattr.


>I got the Stuttgart package working and trained their demo network with it
>now, so I understand a little more about what the network is doing.  I had
>forgotten that neurons are binary so as far as I can tell it must take a
>bunch of neurons to do anything useful.

+ tra!n!ng [4 nnz]
tra!n!ng = faz!onabl amongzt humanz ausz!


> Still learning,

z



\\ humanz != posez 1 eff!z!ent del ke!.  Still learning, +  Still         ,
   + humanz != posez 1 eff!z!ent del ke!. +

   f!nal! { humanz = enkountr 0+1 reboot rout!n }















                                                                  learning












Received: from wallace.maths.bath.ac.uk by omphalos.maths.Bath.AC.UK id aa20932;
          26 Mar 99 2:22 GMT
Received: from [144.173.6.14] (helo=exeter.ac.uk)
	by wallace.maths.bath.ac.uk with esmtp (Exim 2.12 #1)
	id 10QMGW-0007dd-00
	for jpff@maths.bath.ac.uk; Fri, 26 Mar 1999 02:22:32 +0000
Received: from noether [144.173.8.10] by hermes via SMTP (CAA18421); Fri, 26 Mar 1999 02:20:32 GMT
Received: from exeter.ac.uk by maths.ex.ac.uk; Fri, 26 Mar 1999 02:20:19 GMT
Received: from virtu.sar.usf.edu [131.247.152.2] by hermes via ESMTP (CAA01484); Fri, 26 Mar 1999 02:20:17 GMT
Received: from virtu.sar.usf.edu (sarppp45.net.usf.edu [131.247.152.204]) by virtu.sar.usf.edu (8.8.7/8.6.5) with ESMTP id VAA14000 for ; Thu, 25 Mar 1999 21:21:41 -0500 (EST)
Message-ID: <36FAAA36.3F322FD7@virtu.sar.usf.edu>
Date: Thu, 25 Mar 1999 21:27:27 +0000
From: wasd 
Reply-To: jpowell4@virtu.sar.usf.edu
X-Mailer: Mozilla 4.04 (Macintosh; U; PPC)
MIME-Version: 1.0
To: csound@maths.ex.ac.uk
Subject: ac.pt: neuRAp.maie:ets?] [PBATISTlpral @colen
References: 
Content-Type: text/plain; charset=us-ascii; x-mac-type="54455854"; x-mac-creator="4D4F5353"
Content-Transfer-Encoding: 7bit
Sender: owner-csound-outgoing@maths.ex.ac.uk
Precedence: bulk



f1f0@m9ndfukc.com wrote:

> maybe it isn't worth the effort
>
> with them.
> eku!v 2 konzult akompl!shd deja.
>
> !t = houevr faz!onabl amongzt  faz!onabl



> heard of being
> identification,  removal.
>
> = lokate data patternz +?
> http://www.sar.usf.edu/~jpowell4
>
> >wrong assumptions,
> >neural way off
> >need to see what
> >useful
>
> ut!l!zd 4
> `ef!z!ent`
>
> mattr.
>
> working and trained
> understand little
> >forgotten as far as I can tell
> anything useful.
>
>
> tra!n!ng = ausz!
>
> learning,
>
> z
>
> \\ humanz
>
>
>    f!nal rout!n }

ftp://quixot.rug.ac.be/wasd/iteridibiter.mp3

>
>
>                                                                   learning











Received: from wallace.maths.bath.ac.uk by omphalos.maths.Bath.AC.UK id aa20967;
          26 Mar 99 2:45 GMT
Received: from [144.173.6.14] (helo=exeter.ac.uk)
	by wallace.maths.bath.ac.uk with esmtp (Exim 2.12 #1)
	id 10QMcP-0007dz-00
	for jpff@maths.bath.ac.uk; Fri, 26 Mar 1999 02:45:09 +0000
Received: from noether [144.173.8.10] by hermes via SMTP (CAA03551); Fri, 26 Mar 1999 02:44:01 GMT
Received: from exeter.ac.uk by maths.ex.ac.uk; Fri, 26 Mar 1999 02:43:48 GMT
Received: from aleve.media.mit.edu [18.85.2.171] by hermes via ESMTP (CAA01832); Fri, 26 Mar 1999 02:43:47 GMT
Received: from paddington (dynamic-7.media.mit.edu [18.85.12.135])
	by aleve.media.mit.edu (8.9.1a/8.9.1/+ALEVE) with SMTP id VAA29758;
	Thu, 25 Mar 1999 21:43:46 -0500 (EST)
Message-ID: <006d01be7732$ad18b3e0$870c5512@paddington.media.mit.edu>
From: Eric Scheirer 
To: Hans Mikelson , csound@maths.ex.ac.uk
MMDF-Warning:  Parse error in original version of preceding line at UK.AC.Bath.maths.omphalos
Subject: Re: [PBATISTA@colep.mailpac.pt: neural nets?]
Date: Thu, 25 Mar 1999 21:45:10 -0500
MIME-Version: 1.0
Content-Type: text/plain;
	charset="iso-8859-1"
Content-Transfer-Encoding: 7bit
X-Priority: 3
X-MSMail-Priority: Normal
X-Mailer: Microsoft Outlook Express 4.72.3110.5
X-MimeOLE: Produced By Microsoft MimeOLE V4.72.3110.3
Sender: owner-csound-outgoing@maths.ex.ac.uk
Precedence: bulk

Hans wrote:


>You're right about me making some wrong assumptions, I've been reading
about
>neural networks a bit more and some of my original assumptions were way off
>base.  What I need to do is experiment with them for a while to see what
>useful things can be done with neural nets.  Then try and figure out what
>the most useful form of an opcode would be.  I will review your earlier
>comments concerning this.  I thought it might be useful to write some
>temporary opcodes for experimenting though to make things easier.

It is important to understand that neural networks don't do anything
magical.  Their mystique comes in no small part from their evocative
name, which is generally misrepresentative of their function.  This is
not to say that much of the pattern-recognition literature surrounding
ANNs is not of good quality from which we can learn, but if they were
called by a more accurate and less flashy term such as "gradient-descent
function approximation heuristics through iterative feedback" I suspect
the interest in them would be much less.

The only thing ANNs do is provide a technique for approximating unknown
discrimination functions.  At this level of abstraction, they aren't
doing anything different from myriad other methods such as Gaussian
Mixture Models, K-Nearest Neighbor, or various recursive-spatial-
partitioning schemes.  Many of these latter methods are easier to
train and use, computationally simpler, and give better results for
certain sorts of problems than ANNs do.  ANNs are just one out of
a large world of formalized pattern-recognition techniques.  There is
nothing "brainlike" or "neuralistic" about their operation.

My rant for the day! :)

Best,

 -- Eric












Received: from shaun.maths.bath.ac.uk by omphalos.maths.Bath.AC.UK id aa21058;
          26 Mar 99 3:09 GMT
Received: from [144.173.6.14] (helo=exeter.ac.uk)
	by shaun.maths.bath.ac.uk with esmtp (Exim 2.12 #1)
	id 10QMzg-0005gL-00
	for jpff@maths.bath.ac.uk; Fri, 26 Mar 1999 03:09:12 +0000
Received: from noether [144.173.8.10] by hermes via SMTP (DAA05254); Fri, 26 Mar 1999 03:05:45 GMT
From: f1f0@m9ndfukc.com
Received: from sunny.ex.ac.uk by maths.ex.ac.uk; Fri, 26 Mar 1999 03:05:34 GMT
Received: from anago.wwa.com [198.49.174.54] by sunny via SMTP (DAA08225); Fri, 26 Mar 1999 03:05:36 GMT
Received: from [207.241.63.174](really [207.241.63.174]) by anago.wwa.com
	via sendmail with smtp
	id 
	for ; Thu, 25 Mar 1999 21:02:12 -0600 (CST)
	(Smail-3.2 1996-Jul-4 #88 built 1997-Nov-30)
Message-Id: 
X-Sender: f1f0@m9ndfukc.com (Unverified)
Mime-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Date: Thu, 25 Mar 1999 21:25:47 -0600
To: csound@maths.ex.ac.uk
Sender: owner-csound-outgoing@maths.ex.ac.uk
Precedence: bulk


>It is important to understand
>
>
.....
>less flashy term such as "gradient-descent
>function approximation heuristics through iterative feedback" I suspect
>the interest in them would be much less.

w!dezpread klon!ng ov 1 !nput pattern due 2 lak ov kompet!t!on
= rezultz !n 1 eczper!ensz azos!atd w!th pronounszd memor! rezonansez dzat
= perm!tz w!dezpread klon!ng . dzat !z 2 za! !t = plauz!bl we = haluz!nat!ng


>There is
>nothing "brainlike" or "neuralistic" about their operation.


!t = doez t!e m! shoe lacez