From Newsgroup: comp.dsp

Occasionally I attend recitals, where the student
performs a concerto. It's a bit of a financial
stretch for the music department to hire an orchestra,
so a second pianist accompanies, acting as the orchestra.

Thus someone has to condense the entire orchestral section
down to a single instrument! A mighty job of compression
indeed. How do they do that? The conductor isn't a
mathematician, yet manages a feat beyond anything you'd
find in a text.

How would you tackle this one?

--
Rich

--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.dsp

On Tuesday, October 15, 2019 at 11:11:44 AM UTC+13, RichD wrote:

Occasionally I attend recitals, where the student
performs a concerto. It's a bit of a financial
stretch for the music department to hire an orchestra,
so a second pianist accompanies, acting as the orchestra.

Thus someone has to condense the entire orchestral section
down to a single instrument! A mighty job of compression
indeed. How do they do that? The conductor isn't a
mathematician, yet manages a feat beyond anything you'd
find in a text.

How would you tackle this one?

--
Rich

Well you cannot. You just do the main themes.
--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.dsp

<gyansorova@gmail.com> wrote:

On Tuesday, October 15, 2019 at 11:11:44 AM UTC+13, RichD wrote:

Occasionally I attend recitals, where the student
performs a concerto. It's a bit of a financial
stretch for the music department to hire an orchestra,
so a second pianist accompanies, acting as the orchestra.

Thus someone has to condense the entire orchestral section
down to a single instrument! A mighty job of compression
indeed. How do they do that? The conductor isn't a
mathematician, yet manages a feat beyond anything you'd
find in a text.

How would you tackle this one?

A suitably-talented musician arranges the orchestral piece into a piano arrangement.

(This has little to do with the conductor.)

I would believe for major orchestral pieces, such arrangements already
exist and can be purchased. The pianist of course needs to sight-read
the piece, but that part is routine.

Well you cannot. You just do the main themes.

I imagine that is the basis of it.


Steve
--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.dsp

On October 14, Steve Pope wrote:

Occasionally I attend recitals, where the student
performs a concerto...
so a second pianist accompanies, acting as the orchestra.
Thus someone has to condense the entire orchestral section
down to a single instrument! A mighty job of compression
indeed. How do they do that? The conductor isn't a
mathematician, yet manages a feat beyond anything you'd
find in a text.
How would you tackle this one?

A suitably-talented musician arranges the orchestral piece into a piano arrangement.

um, yeah
How does he do that?

I would believe for major orchestral pieces, such arrangements already
exist and can be purchased.

um, yeah
How were they created?

Well you cannot. You just do the main themes.

um, yeah

This is a DSP group, information theory and all that.
The source compression bit splits into a lossless part,
and the more practical, difficult lossy part.

Obviously, condensing an orchestra down to a single
instrument (the piano or violin soloist is fixed) is lossy.
How does one identify the main themes, algorithmically?

I imagine that is the basis of it.

OK, let's go with that - what set of basis functions are
appropriate for this task?

With Fourier spectral analysis, one simply picks the
largest magnitude components. That's a doubtful method, here.

You're given this task as a paid project. You have to
deliver a source compression program...

The point is, music theorists can do this, have done it.
What's their secret sauce?


--
Rich
--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.dsp

RichD <r_delaney2001@yahoo.com> wrote:

On October 14, Steve Pope wrote:

A suitably-talented musician arranges the orchestral piece into a piano
arrangement.

um, yeah
How does he do that?

Study piano from age 6, music theory from age 12, go to a conservatory
and make the grade and they can do that.

Just like musicians would say to us engineers, "how do you do that?"

I would believe for major orchestral pieces, such arrangements already
exist and can be purchased.

um, yeah
How were they created?

See above ... by suitably trained musicians.

This is a DSP group, information theory and all that.

(La-dee-fucking-da)

The source compression bit splits into a lossless part,
and the more practical, difficult lossy part.

Obviously, condensing an orchestra down to a single
instrument (the piano or violin soloist is fixed) is lossy.
How does one identify the main themes, algorithmically?

OK, let's go with that - what set of basis functions are
appropriate for this task?

With Fourier spectral analysis, one simply picks the
largest magnitude components. That's a doubtful method, here.

You're given this task as a paid project. You have to
deliver a source compression program...

I would hire a professional musician, rather than believing that
DSP is some sort of holy grail for all tasks in the world.

Hint: many in the real world are just not down with engineers
believing they can replace musicians. It's bad form at minimum...
some would call us scabs. Musicians who can do this are paid
union scale.


Steve
--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.dsp

RichD <r_delaney2001@yahoo.com> wrote:

Obviously, condensing an orchestra down to a single
instrument (the piano or violin soloist is fixed) is lossy.
How does one identify the main themes, algorithmically?

OK, let's go with that - what set of basis functions are
appropriate for this task?

With Fourier spectral analysis, one simply picks the
largest magnitude components. That's a doubtful method, here.

You're given this task as a paid project. You have to
deliver a source compression program...

One other observation:

For this arranging task, one is not working with the time domain
(audio) signals. One is working in the frequency domain, because musical scores are already "transformed" into the frequency domain,
because they define the amplitude of musical notes. Nor is arranging
a real-time action.

For arranging, the task is score (frequency domain) in, score (frequency domain) out. There is no time-domain signal in the chain.

For performing on an instrument from a score, the task is score
(frequency domain) in, audio (time domain) out.

For scoring/notating, or (most?) compostion, the task is audio
(time domain) in, score (frequency domain) out.

There's a fourth possibility -- time domain in, time domain out.
I will say that corresponds to improvisation, singing in harmony,
and various other things that musicians know how to do.

Steve
--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.dsp

On Mon, 14 Oct 2019 15:11:40 -0700 (PDT), RichD
<r_delaney2001@yahoo.com> wrote:

Occasionally I attend recitals, where the student
performs a concerto. It's a bit of a financial
stretch for the music department to hire an orchestra,
so a second pianist accompanies, acting as the orchestra.

Thus someone has to condense the entire orchestral section
down to a single instrument! A mighty job of compression
indeed. How do they do that? The conductor isn't a
mathematician, yet manages a feat beyond anything you'd
find in a text.

How would you tackle this one?

--
Rich

Have you ever read a synopsis for a movie? Tell the whole story and
all the imagery in a few paragraphs! How do they do that! That's
an extremely high compression rate.


--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.dsp

RichD wrote:

Occasionally I attend recitals, where the student
performs a concerto. It's a bit of a financial
stretch for the music department to hire an orchestra,
so a second pianist accompanies, acting as the orchestra.

Thus someone has to condense the entire orchestral section
down to a single instrument! A mighty job of compression
indeed. How do they do that? The conductor isn't a
mathematician, yet manages a feat beyond anything you'd
find in a text.

How would you tackle this one?

--
Rich

The job is called "arranging". It's "grey art" - a bit art,
a bit science.

You'd be amazed how much you can remove from an arrangement and still
have the piece be recognizable.
--
Les Cargill
--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.dsp

Am 16.10.19 um 02:38 schrieb RichD:

On October 14, Steve Pope wrote:

Occasionally I attend recitals, where the student
performs a concerto...
so a second pianist accompanies, acting as the orchestra.
Thus someone has to condense the entire orchestral section
down to a single instrument! A mighty job of compression
indeed. How do they do that? The conductor isn't a
mathematician, yet manages a feat beyond anything you'd
find in a text.
How would you tackle this one?

A suitably-talented musician arranges the orchestral piece into a piano
arrangement.

um, yeah
How does he do that?

For most parts this is actually quite simple. Your question is posed in
a way as if it were necessary for a symphonic orchestra with, say, 60 musicians, that everyone is always there and plays a different part than
their neighbours. This is not true. You typically have, e.g. 4 second
violins all playing nearly the same thing. When you take this redundancy
out, you'll end up with maybe 8 voices.

And then, still these are not playing in a chaotic way, but European
music has a structure of melody + accompaniment, where the melody can be
a single voice and the accopmaniment plays chords (C major, a minor,
....). For most of the parts in a piece, therefore you could sing the
melody and play the chords on a guitar, to render the piece in a
recognisable way.

On a piano, you can play a few simultaneous voices on one hand and the
chords on the other hand, that's what they do. So you would read the
score, decide which is the dominant melody - often played by the solo musicians and marked up, or played by many in parallel -, extract the
chord from the other notes and replace these second voices by the chord,
which is played in the correct rhythm. With some training it is possible
to do that in "real time", i.e. you know the piece well enough, then you
can play this in a recognisable fashion.

Typically there are a few bars in the piece where this goes wrong, and
one has to think more, but for the majority of the bars you can do that
"by ear".

Christian
--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.dsp

IrCOve seen machine learning demos where you train the network on a large number of similar songs and it generates semi-plausible examples of new songs. I bet the AI community could cook something up that would create piano music from an orchestral arrangement.
Bob
--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.dsp

radams2000@gmail.com writes:

IrCOve seen machine learning demos where you train the network on a
large number of similar songs and it generates semi-plausible examples
of new songs. I bet the AI community could cook something up that
would create piano music from an orchestral arrangement.

I'll bet you're right and I bet 90 percent of the public would be happy
with that. But the 10 percent who had half a musical ear would hate it,
just like they hate lossy codecs.
--
Randy Yates, DSP/Embedded Firmware Developer
Digital Signal Labs
http://www.digitalsignallabs.com
--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.dsp

Randy Yates <yates@digitalsignallabs.com> writes:

radams2000@gmail.com writes:

IrCOve seen machine learning demos where you train the network on a
large number of similar songs and it generates semi-plausible examples
of new songs. I bet the AI community could cook something up that
would create piano music from an orchestral arrangement.

I'll bet you're right and I bet 90 percent of the public would be happy
with that. But the 10 percent who had half a musical ear would hate it,
just like they hate lossy codecs.

I'm on the fence with Sirius XM (just got a trial subscription a couple
months back). There are distinct times when the compression sounds like
&^$*%, but the selection sure is nice.
--
Randy Yates, DSP/Embedded Firmware Developer
Digital Signal Labs
http://www.digitalsignallabs.com
--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.dsp

Randy Yates <yates@digitalsignallabs.com> wrote:

I'm on the fence with Sirius XM (just got a trial subscription a couple >months back). There are distinct times when the compression sounds like >&^$*%, but the selection sure is nice.

Sirius FM is designed to be listened to in vehicles, where you need/want
to listene to a compressed signal. ("Compressed" in the sense of,
listening though a compander set for compression, not in the sense of
bit-rate compression.)

Do they not have an uncompressed option for residential users?

Steve
--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.dsp

spope384@gmail.com (Steve Pope) writes:

Randy Yates <yates@digitalsignallabs.com> wrote:

I'm on the fence with Sirius XM (just got a trial subscription a couple >>months back). There are distinct times when the compression sounds like >>&^$*%, but the selection sure is nice.

Sirius FM is designed to be listened to in vehicles, where you need/want
to listene to a compressed signal. ("Compressed" in the sense of,
listening though a compander set for compression, not in the sense of bit-rate compression.)

Do they not have an uncompressed option for residential users?

Steve

Steve,

I was referring to vehicle listening, and compression in the sense of
lossy signal source compression, not dynamic range compression. While it
may be a necessary evil, that is not something I or anyone else really
wants, but I think most people now just put up with it, or don't even
notice it due to years of this becoming the "norm."
--
Randy Yates, DSP/Embedded Firmware Developer
Digital Signal Labs
http://www.digitalsignallabs.com
--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.dsp

On Sun, 3 Nov 2019 17:14:20 +0000 (UTC), spope384@gmail.com (Steve
Pope) wrote:

Randy Yates <yates@digitalsignallabs.com> wrote:

I'm on the fence with Sirius XM (just got a trial subscription a couple >>months back). There are distinct times when the compression sounds like >>&^$*%, but the selection sure is nice.

Sirius FM is designed to be listened to in vehicles, where you need/want
to listene to a compressed signal. ("Compressed" in the sense of,
listening though a compander set for compression, not in the sense of >bit-rate compression.)

Do they not have an uncompressed option for residential users?

Steve

They have a ton of channels now that aren't broadcast, but available
streaming on-line only. I think some of them may be free to
non-subscribers, but you can check on their website if interested.

I've no idea what level of compression they apply or don't to the
on-line streaming stuff, but I'd be surprised if they ever send
uncompressed content just because it would be a waste of BW.




--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.dsp

Eric Jacobsen <theman@ericjacobsen.org> wrote:

Sirius FM is designed to be listened to in vehicles, where you need/want
to listene to a compressed signal. ("Compressed" in the sense of, >>listening though a compander set for compression, not in the sense of >>bit-rate compression.)

Do they not have an uncompressed option for residential users?

They have a ton of channels now that aren't broadcast, but available >streaming on-line only. I think some of them may be free to >non-subscribers, but you can check on their website if interested.

I've no idea what level of compression they apply or don't to the
on-line streaming stuff, but I'd be surprised if they ever send
uncompressed content just because it would be a waste of BW.

I may not have been clear but -- "compressed" in that the audio is
possibly compressed, not in that the bit-stream is compressed (which
it clearly must be).

Steve
--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.dsp

On Mon, 11 Nov 2019 18:34:13 +0000 (UTC), spope384@gmail.com (Steve
Pope) wrote:

Eric Jacobsen <theman@ericjacobsen.org> wrote:

Sirius FM is designed to be listened to in vehicles, where you need/want >>>to listene to a compressed signal. ("Compressed" in the sense of, >>>listening though a compander set for compression, not in the sense of >>>bit-rate compression.)

Do they not have an uncompressed option for residential users?

They have a ton of channels now that aren't broadcast, but available >>streaming on-line only. I think some of them may be free to >>non-subscribers, but you can check on their website if interested.

I've no idea what level of compression they apply or don't to the
on-line streaming stuff, but I'd be surprised if they ever send >>uncompressed content just because it would be a waste of BW.

I may not have been clear but -- "compressed" in that the audio is
possibly compressed, not in that the bit-stream is compressed (which
it clearly must be).

Steve

If transmitting to different devices I don't know why that wouldn't
just be handled at the terminal device. Keeping the dgitized signal
linear allows use at any device with a little bit of processing at the
device to suit that device or installation. Not sure why companding
would be distributed in a digital transmission.




--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.dsp

Eric Jacobsen <theman@ericjacobsen.org> wrote:

If transmitting to different devices I don't know why that wouldn't
just be handled at the terminal device. Keeping the dgitized signal
linear allows use at any device with a little bit of processing at the
device to suit that device or installation. Not sure why companding
would be distributed in a digital transmission.

Certainly. OP had compained about the "compressed" audio though...
worth disambiguating what was going on.

Steve
--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.dsp

Steve Pope <spope384@gmail.com> wrote:

Eric Jacobsen <theman@ericjacobsen.org> wrote:

If transmitting to different devices I don't know why that wouldn't
just be handled at the terminal device. Keeping the dgitized signal >>linear allows use at any device with a little bit of processing at the >>device to suit that device or installation. Not sure why companding
would be distributed in a digital transmission.

Certainly. OP had compained about the "compressed" audio though...
worth disambiguating what was going on.

Another aspect - companding at the endpoint may not be sufficient.

"Radio ready" mixes are often not simply compressed, but tweaked to
the source material by engineers.

https://www.sageaudio.com/blog/pre-mastering-tips/mixing-radio.php


Steve
--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.dsp

On Fri, 15 Nov 2019 13:12:39 +0000 (UTC), spope384@gmail.com (Steve
Pope) wrote:

Steve Pope <spope384@gmail.com> wrote:

Eric Jacobsen <theman@ericjacobsen.org> wrote:

If transmitting to different devices I don't know why that wouldn't
just be handled at the terminal device. Keeping the dgitized signal >>>linear allows use at any device with a little bit of processing at the >>>device to suit that device or installation. Not sure why companding >>>would be distributed in a digital transmission.

Certainly. OP had compained about the "compressed" audio though...
worth disambiguating what was going on.

Another aspect - companding at the endpoint may not be sufficient.

"Radio ready" mixes are often not simply compressed, but tweaked to
the source material by engineers.

https://www.sageaudio.com/blog/pre-mastering-tips/mixing-radio.php

Steve

It does drive me a little bit bonkers when two tunes play
back-to-back, on SiriusXM or broadcast radio or whatever, and a
significant volume adjustment is required because of the differences
in mixing. Kind of like how they used to predistort commercials to
make them seem louder. So not everybody is doing this uniformly and
it's probably just annoying people. ;)

Back in the day there were standards for the analog deviation on FM
signals and in the FM audio subcarrier for NTSC TV signals. Maybe
there still is, as our local NPR FM station always requires adjusting
the volume up a lot more than most other broadcast stations, and
public broadcast has historically been notorious for sticking to
published regs...i.e., everybody else probably turns up their
deviation a bit.

When I was in grad school I took care of some of the engineering for a
local low-power TV station that swapped out an (illegal) modulator for
a more kosher one. The equipment needed to set the audio deviation
per the technical regs wasn't available so I just took a little
portable TV with me to the transmitter shack and adjusted the
deviation until it sounded about like everybody else, i.e., I didn't
have to adjust the volume when I tuned between it and other stations.

So it's not that hard, and it just seems annoying when it gets
overdone.

/End grumpy rant.

--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.dsp

Eric Jacobsen <theman@ericjacobsen.org> wrote:

On Fri, 15 Nov 2019 13:12:39 +0000 (UTC), spope384@gmail.com (Steve

Pope) wrote:

Another aspect - companding at the endpoint may not be sufficient.

"Radio ready" mixes are often not simply compressed, but tweaked to
the source material by engineers.

https://www.sageaudio.com/blog/pre-mastering-tips/mixing-radio.php

It does drive me a little bit bonkers when two tunes play
back-to-back, on SiriusXM or broadcast radio or whatever, and a
significant volume adjustment is required because of the differences
in mixing. Kind of like how they used to predistort commercials to
make them seem louder. So not everybody is doing this uniformly and
it's probably just annoying people. ;)

Back in the day there were standards for the analog deviation on FM
signals and in the FM audio subcarrier for NTSC TV signals. Maybe
there still is, as our local NPR FM station always requires adjusting
the volume up a lot more than most other broadcast stations, and
public broadcast has historically been notorious for sticking to
published regs...i.e., everybody else probably turns up their
deviation a bit.

When I was in grad school I took care of some of the engineering for a
local low-power TV station that swapped out an (illegal) modulator for
a more kosher one. The equipment needed to set the audio deviation
per the technical regs wasn't available so I just took a little
portable TV with me to the transmitter shack and adjusted the
deviation until it sounded about like everybody else, i.e., I didn't
have to adjust the volume when I tuned between it and other stations.

Yes, it does seem the rules re. FM deviation (i.e. you couldn't
undermodulate) have been abandonded. I still listen to a lot of FM
radio -- mostly college radio -- and they feel free to have very
quiet sections of audio. Also, one no longer is required to have
at least a 3rd Class Radiotelephone Engineer on site when broadcasing.

But for vehicular listening you do want to compress the dynamic
range. I hear casettes are making a comeback. Back then, if making
cassettes to listen to in the car, I'd just turn on the limiter
and shove the input levels away up. Primitive technology works.

Steve
--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.dsp

On Tuesday, October 15, 2019 at 7:57:58 PM UTC-7, Steve Pope wrote:

(snip)

I would hire a professional musician, rather than believing that
DSP is some sort of holy grail for all tasks in the world.

Hint: many in the real world are just not down with engineers
believing they can replace musicians. It's bad form at minimum...
some would call us scabs. Musicians who can do this are paid
union scale.

Many musicians I have known are also interested in math or physics
or engineering, so I suspect that you wouldn't have such a hard
time finding someone interested in both.

Reminds me, though, of Shazam:

https://www.shazam.com/

Shazam will identify a musical recording given a small sample
of an often noisy version of the source. Not so obvious, it makes
an exact match to the original. That is, it won't just identify
the musical piece but the exact CD that it came from.

They do this by extracting some parts of the signal, certain
frequency ranges, fingerprint those, and then compare them with
saved fingerprints. That is, they have the ability to compress
a musical composition down to a fairly small number of bits,
and match up those bits.
--- Synchronet 3.22a-Linux NewsLink 1.2