adaptive tuning. Can a computer pick a melody from the harmony ?

Hello tuners,
I have been recently thinking about an answer to real time adaptive tuning which
I have almost solved.

There is one problem though.

Is it possible for a computer to identify the notes of a melody. That is to pick
out notes of a melody from the simultaneously sounding harmonies.

To do this we need to catalogue the characteristics of a melody. We need to
find out those things that, in particular, differentiate a melody from harmonic
support.

Firstly there is tessitura, or the range in which the pitches are played. These
genrally conform to a fairly restricted range for most melodies. Generally two
octaves is sufficient to play most melodies. This then is one way that a
computer could narrow down the choices of sounding notes to decide on the melody
note.

But of course it is quite likely that there will be harmony of some sort in the
same tessitura as the melody.

Are there any other ways that a melody differs from harmony?

Perhaps the proportion between the speed of movement of the melody and the
harmony. Generally speaking again melodies move faster in proportion to
melodies. This is something a computer could also measure.

However these characteristics are hardly enough. There must be more elements to
melodies that could be noticed by a computer.

Any ideas?

Justin White

[Justin White wrote:]
>I have been recently thinking about an answer to real time adaptive
>tuning which I have almost solved.

Please tell us more!

>There is one problem though.

>Is it possible for a computer to identify the notes of a melody. That
>is to pick out notes of a melody from the simultaneously sounding
>harmonies.

I think it is the exception for there to be a clear distinction between
melody and harmony, at least in the music I listen to and tune.

>To do this we need to catalogue the characteristics of a melody. We
>need to find out those things that, in particular, differentiate a
>melody from harmonic support.

>Firstly there is tessitura, or the range in which the pitches are
>played. These genrally conform to a fairly restricted range for most
>melodies. Generally two octaves is sufficient to play most melodies.
>This then is one way that a computer could narrow down the choices of
>sounding notes to decide on the melody note.

So much music bounces motifs among many voices, some high and some low.

>But of course it is quite likely that there will be harmony of some
>sort in the same tessitura as the melody.

Absolutely.

>Are there any other ways that a melody differs from harmony?

>Perhaps the proportion between the speed of movement of the melody and
>the harmony. Generally speaking again melodies move faster in
>proportion to melodies.

You mean, in proportion to harmonies? That is often true some of the
time, but it can suddenly change, too.

>This is something a computer could also measure.

>However these characteristics are hardly enough. There must be more
>elements to melodies that could be noticed by a computer.

>Any ideas?

At the risk of being Captain Bringdown, it doesn't seem to me to be
theoretically possible to make such a distinction. But if it IS
possible, what would your method do with the information? That is,
how does whether a note is melody vs. harmony affect the way you would
adaptively tune it?

JdL

--- In tuning@y..., "John A. deLaubenfels" <jdl@a...> wrote:
> [Justin White wrote:]
> >I have been recently thinking about an answer to real time
adaptive
> >tuning which I have almost solved.
>

I am thinking about Indian music, which is only melody: Is the method
for adaptive training ready to be put to use?

Haresh.

--- In tuning@y..., "Haresh BAKSHI" <hareshbakshi@h...> wrote:
> --- In tuning@y..., "John A. deLaubenfels" <jdl@a...> wrote:
> > [Justin White wrote:]
> > >I have been recently thinking about an answer to real time
> adaptive
> > >tuning which I have almost solved.
> I am thinking about Indian music, which is only melody: Is the
method for adaptive training ready to be put to use?>
> Haresh. >>>>

I meant adaptive TUNING.

Haresh.

--- In tuning@y..., "Haresh BAKSHI" <hareshbakshi@h...> wrote:
> --- In tuning@y..., "Haresh BAKSHI" <hareshbakshi@h...> wrote:
> > --- In tuning@y..., "John A. deLaubenfels" <jdl@a...> wrote:
> > > [Justin White wrote:]
> > > >I have been recently thinking about an answer to real time
> > adaptive
> > > >tuning which I have almost solved.
> > I am thinking about Indian music, which is only melody: Is the
> method for adaptive training ready to be put to use?>
> > Haresh. >>>>
>
> I meant adaptive TUNING.
>
> Haresh.

Hi Haresh.

John's adaptive tuning software _could_ work for Indian music,
because Indian music has a constant drone. However, the result, if
I'm not mistaken, would always be a _fixed_ tuning rather than an
adaptive tuning (if we're not including 9-limit intervals). If the
drone is C-G, for example, you'd get the following 5-limit tuning:

C C# D Eb E F G Ab A Bb B C
1/1 ? 9/8 6/5 5/4 4/3 3/2 8/5 5/3 9/5 15/8 2/1

Note that this agrees with the "Modern Indian Gamut" except for two
points:

1) A is 5/3 instead of 27/16
2) C# is not well-defined (actually, in my interpretation of the
Modern Indian Gamut, it's also the only note that's not well-defined,
serving as both 16/15 and 135/128).

--- In tuning@y..., paul@s... wrote:

>>>>> > Hi Haresh.>
> John's adaptive tuning software _could_ work for Indian music,
> because Indian music has a constant drone. However, the result, if
> I'm not mistaken, would always be a _fixed_ tuning rather than an
> adaptive tuning (if we're not including 9-limit intervals). If the
> drone is C-G, for example, you'd get the following 5-limit tuning:
>
> C C# D Eb E F G Ab A Bb B C
> 1/1 ? 9/8 6/5 5/4 4/3 3/2 8/5 5/3 9/5 15/8 2/1
>>>>>>>

Hi Paul, thanks for your input. The question of A, shown above, is
tricky. I would say that any such cases affecting the Indian gamut by
even close to 2 cents, would create problems. The C# situation would
be "clear", one way or the other.

In addition to the C-G tanpura tuning you treat of, above, we have
two other variations: C-F, and, in case of very few raga-s, C-B. I
would appreciate you comments, especially how the C-B type would be
handled.

Regards,
Haresh.

--- In tuning@y..., "Haresh BAKSHI" <hareshbakshi@h...> wrote:

> Hi Paul, thanks for your input. The question of A, shown above, is
> tricky. I would say that any such cases affecting the Indian gamut
by
> even close to 2 cents, would create problems.

Meaning you wouldn't accept a deviation, even a very small one, from
the accepted Gamut ratios?

Oddly, I left F# out of my post. This would receive another "?",
despite its specification as 45/32 in the Modern Indian Gamut.

> In addition to the C-G tanpura tuning you treat of, above, we have
> two other variations: C-F,

For which the results would simply be the same as the C-G results,
transposed down a fifth.

> and, in case of very few raga-s, C-B.

If you somehow made sure the drone was tuned 15:8, then I get the
following:

C C# D Eb E F F# G Ab A Bb B C
1/1 ? 9/8 32/27 5/4 4/3 45/32 3/2 128/81 5/3 ? 15/8 2/1
* *

*For Eb and Ab the program will seek out a compromise between two
conflicting ratios.

Eb will compromise between 6/5 and 75/64 (32/27 is only 1 cent from
the middle).

Ab will compromise between 8/5 and 25/16 (128/81 is only 1 cent from
the middle).

This is an exciting new interpretation of these Pythagorean shrutis
that I have not come across before.

--- In tuning@y..., paul@s... wrote:
> --- In tuning@y..., "Haresh BAKSHI" <hareshbakshi@h...> wrote:
>
> > Hi Paul, thanks for your input. The question of A, shown above,
is tricky. I would say that any such cases affecting the Indian
gamut by even close to 2 cents, would create problems. >>>>>>>

>>> Meaning you wouldn't accept a deviation, even a very small one,
from the accepted Gamut ratios? >>>

Yes, though I may sound a little theoretical. I feel that such a
deviation would affect the melody far too much -- it would, perhaps,
affect harmony even more, is it?

> This is an exciting new interpretation of these Pythagorean shrutis
> that I have not come across before. >>

Indeed, this has taken an exciting turn. Thanks.

Haresh.

--- In tuning@y..., paul@s... wrote:

/tuning/topicId_22314.html#22346

> This is an exciting new interpretation of these Pythagorean
> shrutis that I have not come across before.

Spoken like a true die-hard tuning theorist.

... it made me laugh the first time I read it.

Jacky, isn't there another Broadway song in here somewhere?
I can almost here the tune under it... try it: read Paul's
words again as a vocal line, with Broadway music going along
with it.

(Uh oh... looks like the first intentionally microtonal
Broadway musical is shaping up!...)

-monz
http://www.monz.org
"All roads lead to n^0"

>>I have been recently thinking about an answer to real time adaptive
>>tuning which I have almost solved.

>Please tell us more!

Hello John,

What I envisage is based on the research of Boomsliter and Creel who came up
with a theory for melodic intonation in the 1960's. I'll post you a copy of
their paper if you are interested in learning more.

Basically their theory is that the human auditary system finds melodies in small
whole number ratios too simple for any enjoyment. So rather than selecting notes
that are directly related to DO or 1/1 we hear melodies tuned [in just ratios]
to chains of reference.

They propose 12 basic ratios 1/1, 9/8, 7/6, 6/5, 5/4, 4/3, 7/5, 3/2, 8/5, 5/3,
7/4, 9/5.
These ratios are transposed 4 times [sometimes more] in three different schemes.
What they term a major linkage which transposes the previous basic scale by a
chain of 3/2's, a minor linkage that transposes the same scale by an initial 6/5
and a blue reference chain that begins with an initial 5/4 and then a further 3
transpositions by 3/2.

They came upon this melodic matrix by empirical testing of musicians. That is
the theory was their logical organisation of the data retrieved through
experiments.

The way this matrix works for melodic intonation is a melody will begin using
the above scale and then as the melody progresses the the reference changes
going further away from the initial 1/1. This provides the drive towards
resolution at the end of the piece. It should be emphasised that there is no
change of key. Often ones from different references are played against each
other. To chage key this whole matrix would need to be transposed to a new
initial DO.

What I am proposing is an electronic keyboard that uses two pedals [or pitch
wheel with stops on each reference] to move forwards and backwards in the chain
of reference. There would be a selector button to choose either the major,
minor, or blue reference. Also there would be a harmonic duodene array above the
normal keyboard. This duodene would be used to change keys. Although I hazard
that real keychanges are far less common in music than we think [I feel most
keychanges can be accomplished without retuning the whole matrix. This way there
is drive for reslolution on the initial 1/1].

The melody in this system would be the unchanging factor. Chords would
adaptively tune around the melody note [similar to babershop where the lead does
the least adptive tuning].

I haven't exactly worked out what notes the chords will retune to but it will be
context specfic. It will depend on what other notes are currently sounding. For
instance if C, E, G & B are sounding as a chord the tuning will be
1/1,5/4,3/2,15/8 but if say the chord is C, Eb, G, B then it will be tuned
1/1, 6/5, 3/2, 9/5.

>>There is one problem though.

>>Is it possible for a computer to identify the notes of a melody. That
>>is to pick out notes of a melody from the simultaneously sounding
>>harmonies.

>I think it is the exception for there to be a clear distinction between
>melody and harmony, at least in the music I listen to and tune.

>>To do this we need to catalogue the characteristics of a melody. We
>>need to find out those things that, in particular, differentiate a
>>melody from harmonic support.

>>Firstly there is tessitura, or the range in which the pitches are
>>played. These genrally conform to a fairly restricted range for most
>>melodies. Generally two octaves is sufficient to play most melodies.
>>This then is one way that a computer could narrow down the choices of
>>sounding notes to decide on the melody note.

>So much music bounces motifs among many voices, some high and some low.

>>But of course it is quite likely that there will be harmony of some
>>sort in the same tessitura as the melody.

>Absolutely.

>>Are there any other ways that a melody differs from harmony?

>>Perhaps the proportion between the speed of movement of the melody and
>>the harmony. Generally speaking again melodies move faster in
>>proportion to melodies.

>You mean, in proportion to harmonies? That is often true some of the
>time, but it can suddenly change, too.

Yes sorry for the mistake.

>>This is something a computer could also measure.

>>However these characteristics are hardly enough. There must be more
>>elements to melodies that could be noticed by a computer.

>>Any ideas?

>At the risk of being Captain Bringdown, it doesn't seem to me to be
>theoretically possible to make such a distinction. But if it IS
>possible, what would your method do with the information? That is,
>how does whether a note is melody vs. harmony affect the way you would
>adaptively tune it?

While reading your post I have though up two solutions to the problem of
separating melody from harmony. One is to have two keyboards [lefthand:
harmony, righthand:melody] on one keyboard you would play the melody while on
the other you would play the harmony. The melody keyboard would be monophonic
and it's output would provide the basis for the polyphonic harmony keyboard to
adaptively tune to. The melody keyboard could be either a standard keyboard with
the reference chains operated via pedals or could be a Bosanquet style keyboard
so that all the melody notes could be at hand.

Haresh wrote:

<I am thinking about Indian music, which is only melody: Is the method
<for adaptive training ready to be put to use?

No at this stage I'm only thinking about how to make it work the way I want. So
at this stage there is no software or little black box. But once I get some
feedback I'll start on a prototype synth. I just have to work out if there any
tuning theory problems that need to be ironed out first.

There are however 4 software programs that do adaptive tuning to some degree.
They are are very own John de Laubenfels JIRelay [windows], Justonics Pitch
Palette [windows & mac], Realtime tuner [mac only] and another one which is a
softsynth. Ican't remember the name though.

My method as proposed should work with Indian music. But Haresh what did you
want the software or keyboard to do ? As I see it there would be two ways of
approaching Indian music using adaptive tuning. One would be two keep the melody
static and have harmonies adaptively tuning around it or you could allow the
scale to retune in relation to different tonics. All the products listed
previously can do the latter. I know JIRelay and Realtime tuner can do the
former but I am unsure as to the other two.

Justin

--- In tuning@y..., "Justin White" <justin.white@d...> wrote:

/tuning/topicId_22314.html#22356

> What I envisage is based on the research of Boomsliter and
> Creel who came up with a theory for melodic intonation in
> the 1960's. I'll post you a copy of their paper if you are
> interested in learning more.
>
> Basically their theory is that the human auditary system finds
> melodies in small whole number ratios too simple for any
> enjoyment. So rather than selecting notes that are directly
> related to DO or 1/1 we hear melodies tuned [in just ratios]
> to chains of reference.

Hi Justin,

For the record, thought I'd let you know that I think B&C's
ideas merit much more attention, and I'm glad you want to
create software that makes use of them.

I've mentioned my (currently stalled) JustMusic project here
a number of times in the last few days. Take a look and see
if you'd like to be a part of it:
/justmusic

It would probably be very easy to incorporate your ideas into
JustMusic rather than do it all from scratch.

-monz
http://www.monz.org
"All roads lead to n^0"

--- In tuning@y..., "Haresh BAKSHI" <hareshbakshi@h...> wrote:
> --- In tuning@y..., paul@s... wrote:
> > --- In tuning@y..., "Haresh BAKSHI" <hareshbakshi@h...> wrote:
> >
> > > Hi Paul, thanks for your input. The question of A, shown above,
> is tricky. I would say that any such cases affecting the Indian
> gamut by even close to 2 cents, would create problems. >>>>>>>
>
> >>> Meaning you wouldn't accept a deviation, even a very small one,
> from the accepted Gamut ratios? >>>
>
> Yes, though I may sound a little theoretical. I feel that such a
> deviation would affect the melody far too much -- it would,
perhaps,
> affect harmony even more, is it?

While the 21.5 cent difference between A 5/3 and A 27/16 is enough to
affect melody and more than enough to affect harmony, a 2 cent
difference on acoustic Indian instruments . . . ?
>
>
> > This is an exciting new interpretation of these Pythagorean
shrutis
> > that I have not come across before. >>
>
> Indeed, this has taken an exciting turn. Thanks.

Thanks for continuing to probe.

--- In tuning@y..., "Justin White" <justin.white@d...> wrote:
>
>
>>>> But Haresh what did you want the software or keyboard to do ? As
I see it there would be two ways of approaching Indian music using
adaptive tuning. One would be two keep the melody static and have
harmonies adaptively tuning around it or you could allow the scale to
retune in relation to different tonics. All the products listed
previously can do the latter. I know JIRelay and Realtime tuner can
do the former but I am unsure as to the other two. >>>>

Hi Justin, actually, BOTH. The first, keeping the melody static and
have harmonies adaptively tuning around it. My experiments and
meanderings in the genre which has come to be called "World Music",
show that this has great potential for introducing JI in this genre.

The second, where you could allow the scale to retune in relation to
different tonics, has, again, a very significant application in the
field of research in Indian classical music. This takes me directly
to "extended reference" [B & C]. I suspect that we have been using
extended reference, by dwelling on the vadi, and then samvadi, and
weaving patterns around them, during the raga improvisation. This
would be very much like "in relation to different tonics" that you
refer to, above. I have had a little offlist exchange of views with
Paul on this topic. Going back and forth between the tonic "Sa", and
the "new tonic", the vadi/samvadi, may throw new light on how "rasa"
(aesthetic emotions) works in Indian classical music.

Your comments, please?
Regards,
Haresh.

Hello Kraig ,

>Justin!
> I am likewise a big B&C fan
>see below

Good to hear it !

>Justin White wrote:

>>
>> What I am proposing is an electronic keyboard that uses two pedals [or pitch
>> wheel with stops on each reference] to move forwards and backwards in the
chain
>> of reference. There would be a selector button to choose either the major,
>> minor, or blue reference. Also there would be a harmonic duodene array above
the
>> normal keyboard. This duodene would be used to change keys. Although I hazard
>> that real keychanges are far less common in music than we think [I feel most
>> keychanges can be accomplished without retuning the whole matrix. This way
there
>> is drive for reslolution on the initial 1/1].
>>
>> The melody in this system would be the unchanging factor. Chords would
>> adaptively tune around the melody note [similar to babershop where the lead
does
>> the least adptive tuning].

>Interesting. and will enjoy hearing the results. It seems to me that these
extended >references
>might be what lead to the impulses of modulating in the first place. if so, one
would >tend to
>think that the adaptation route would water down the dynamics of the melody.

Absolutely ! The inclusion of the duodene method of key change as you have
rightly noted would be pointless. To retune the whole referential matrix would
make the new key's 1/1 just as restful and resolved as the previous key center.

I thought of including this option only because other people might want it. For
myself I would never use it.

I prusume this is what you mean ? Beacause I don't think the barbershop method
of bending chords to make them just in relation to the currently sounding melody
would be in any way detriment to the dynamics of the melody. I say this because
used in this way the B&C melody system would not change at all it would be the
rock around which the waters of harmony would bend.

> It seems one could
>let the harmony be based on the same extended reference. Not saying you are
right or >wrong but
>just wanting to know how you ended up drawn to using it his way.

I have been using Justonics pitch palette software to achieve B&C's reference
matrix
.With this software it is easy to use a retuning octace for keychange [used for
cycling between different references] and another octave for choosing the tonic
to which the chords sound. Unfortunatly sounding notes cannot be retuned and I
am restricted to the notes of the B&C melody scale for harmony. Major 7th can't
be tuned. I wanted a separation between harmony and melody and the ability to
retune sounding notes to achieve just chords all of the time. The only way I can
accomplish this is with a hardware solution. No existing synth or sound card can
accomplish this. I have thought of rewriting the operating system for my Ensoniq
ASR-X as it does support the midi tuning standard [but not the retuning of
sounding notes].

>Erv's 17 tone scale has dealt with
>in
> http://www.anaphoria.com/genus.PDF is his own solution to this problem.
>Erv did so layouts labeled Duodene but can't seem to find it right now. will
keep >looking and it
>might already be up

Yes I was attracted to this scale. I thought of creating a scale in the smae
manner using a septimal tetrachord...I haven't found a tetrachord that will give
me the tetrad s I want yet.

>-- Kraig Grady
>North American Embassy of Anaphoria island
> http://www.anaphoria.com

>The Wandering Medicine Show
>Wed. 8-9 KXLU 88.9 fm

Justin White

--- In tuning@y...,
/tuning/topicId_22314.html#22356

I wrote :

>> What I envisage is based on the research of Boomsliter and
>> Creel who came up with a theory for melodic intonation in
>> the 1960's. I'll post you a copy of their paper if you are
>> interested in learning more.
>>
>> Basically their theory is that the human auditary system finds
>> melodies in small whole number ratios too simple for any
>> enjoyment. So rather than selecting notes that are directly
>> related to DO or 1/1 we hear melodies tuned [in just ratios]
>> to chains of reference.

Monz wrote:

>Hi Justin,

>For the record, thought I'd let you know that I think B&C's
>ideas merit much more attention, and I'm glad you want to
>create software that makes use of them.

>I've mentioned my (currently stalled) JustMusic project here
>a number of times in the last few days. Take a look and see
>if you'd like to be a part of it:
>/justmusic

>It would probably be very easy to incorporate your ideas into
>JustMusic rather than do it all from scratch.

Thanks for the offer Monz, I'm already a subscriber to your JustMusic list.
However what I am proposing is not currently possible to achieve with software.
A new soundcard would have to be made to achieve retuning of sounding notes
without using midi pitch bend. I'm more interested in the hardware aspects. But
I'd like to be able to integrate some hardware editing features into your
program. And a software interface for it would be a great thing to integrate
with JustMusic.

Incidently, I thought of your use of 75/64 in "A close shave" as possibly being
a 5:4 relation to 25/16 which may have been the refence at that moment. I would
have to find out what other absolute ratios were in close proximity to this one
to say conclusively what level of reference this note is operating on.

Justin White

--- In tuning@y..., "Justin White" <justin.white@d...> wrote:
>
> Yes I was attracted to this scale. I thought of creating a scale in
the smae
> manner using a septimal tetrachord...I haven't found a tetrachord
that will give
> me the tetrad s I want yet.

Can you explain what you're trying to do? Maybe I can help.