Was this aimed at ASR?

[Mort]

This looks like elaborate semantic quibbling to me.

This whole recent thread qualifies as such.

 

[rdenney]

Playing music is not the same as composing it. Each time somebody plays a composition they are reproducing it. Joni Mitchell commented on this. She said if someone paints a great piece of art, no one ever askes them, "Hey, paint me Starry Night again!" She liked painting because she only had to do a painting once. Now a copy of a painting made by a machine may have some constraints. Will it try to copy 3D effects, like thickness of paint in certain areas? Will it try to copy the canvas material properties so it feels the same? Will it try to scale to the exact same size, or blow it up bigger, or smaller? Will it use a reduced color set, maybe just black and white? Are these purely engineering decisions, or is it somewhat of a blend? Maybe there are engineering constraints that require some artful decisions to get a good looking result.

That’s not how it works. The notated music is simply not fully dispositive, any more than written language determines one’s choices of accent, emphasis, and phrasing when read aloud by an actor. Even when following the notated instructions strictly, the performer has a lot of room for interpretation, and permission from the composer to apply it. Performance is part of production, not reproduction. No composer thinks their score is a complete product without it being performed.

The components of a playback system, on the other hand, are not artists and are not given incomplete instructions with the permission to further interpret them as they see fit. A recorded signal is a comprehensively described waveform providing everything the playback system needs to reproduce it accurately. When playback components apply their own interpretation, the results may be pleasing to some and offensive to others. But even if they are pleasing because they correct a fault in the way they were recorded, they impose a departure from the artists’ intentions, and the formula is just as likely to undermine the next recording as enhance it.

This is a fundamental philosophy of most on this forum. Many in the audio press oppose it, leading to generations of enthusiasts chasing the next ridiculous tweak, over and over again, spiraling without controls into a greater distance from the recordings they profess to love.

By the way, musicians are quite often the least qualified to understand this. If, as a musician, I thought a cascade of subjective evaluations and tweaks optimized my recordings beyond what was laid down, I’d fire my producer.

Rick “not that anyone produces measurable examples of these ‘enhancements’” Denney

 

[Tim Link]

So that’s the recording engineer you are speaking about, not the audio components.

That's how it's been traditionally. We may reach a point where playback and recording engineering get blurred. Back in the 1980s I was already debating with my friends how a home audio system should work. I argued that the system's job is to accurately reproduce the sound on the recording. My friends felt that audio systems should sound unique, have their own character. As an end user they wanted to have some say in the artistic endeavor, not just be an end consumer.

 

[Tim Link]

That’s not how it works. The notated music is simply not fully dispositive

Your point is understood. I think previous comments made by others and myself cover your argument thoroughly enough that I have nothing novel to add for now.

I want you ask you though what you think of something like crosstalk reduction methods such as BACCH. Is this just another silly tweak or a serious correction? There's an argument that the recording was never meant to be heard that way. But at the same time crosstalk nulls in a phantom image created by two speakers prevent an accurate portrayal of the original sound that was being recorded. Crosstalk reduction can remove the nulls, but at the same time ends up making the stereo imaging wider than anticipated. What's the correct approach?

 

[Jim Taylor]

My friends felt that audio systems should sound unique, have their own character. As an end user they wanted to have some say in the artistic endeavor, not just be an end consumer.

Sounds like petty ego.

 

[Multicore]

Ok I can follow your pedantry but consider that reproduce means 1. produce a copy or 2. produce something very similar to (something else) in a different medium or context.

The latter is pretty much what performing a musical score/composition/set of instruction is.

From my experience as a musician who sometimes plays compositions and uses improvisation, as almost all musicians do outside the classical world, I can't really get behind this idea.

Recently I've been working towards making a recording of Ghosts by Albert Ayler. It is, without a shadow of doubt or possible controversy, the greatest tune to emerge in the 20th century and Don Cherry was right to insist that it should be the national anthem of the USA.

I can reproduce this, i.e. play it dead straight according to the score (a useful step in learning it) but that's an absurd thing to do. Afaict from the half dozen recordings of Ayler I've studied, he never reproduced it. What did Ayler and his bands do instead of reproduction? The verbs I find useful here are to perform and to interpret, both of which contrast with to reproduce. Since I'm playing it on guitar I also have to arrange it, a useful technical term.

Specifically for my recording, I want to imagine how Sonny Sharrock might have interpreted it, in a counterfactual history in which it would be on his album Guitar. Would Sonny Sharrock have reproduced Ghosts?

Noel Akchote has five different interpretations and arrangements of Ghosts in his collection last year Of Albert (Complete Plays Ayler, 2025). None of those seem like reproductions to me. What do you think of this delightful version, under 1-minute long?

Ghosts (Arr. Gary Windo, 1988), by Noël Akchoté

from the album Of Albert (Complete Plays Ayler, 2025)

noelakchote.bandcamp.com

N.A.D.'s 1989 Italian recording with Fred Frith on violin is high on my list too. Sharrock appears later on that album providing the solo to Bibbidi-Bobbidi-Boo (yes, that one).

This morning I had the idea to grow a manouch, don a fruity cravate, and do a Le Jazz Hot-style version of Ghosts.

Is it even possible to reproduce a composition like Ghosts? Yes but no.

 

[ahofer]

That's how it's been traditionally. We may reach a point where playback and recording engineering get blurred. Back in the 1980s I was already debating with my friends how a home audio system should work. I argued that the system's job is to accurately reproduce the sound on the recording. My friends felt that audio systems should sound unique, have their own character. As an end user they wanted to have some say in the artistic endeavor, not just be an end consumer.

The interesting thing about this is that most often the same people who say this eschew the most flexible and inexpensive methods of accomplishing their goals (EQ, plugins, etc.), pursue the inflexible, indirect, and expensive methods (e.g. seeking 'synergies' among expensive, colored high end audio), and after all that they generally can't hear 80% of the effects they claim to achieve.

 

[rdenney]

Your point is understood. I think previous comments made by others and myself cover your argument thoroughly enough that I have nothing novel to add for now.

I want you ask you though what you think of something like crosstalk reduction methods such as BACCH. Is this just another silly tweak or a serious correction? There's an argument that the recording was never meant to be heard that way. But at the same time crosstalk nulls in a phantom image created by two speakers prevent an accurate portrayal of the original sound that was being recorded. Crosstalk reduction can remove the nulls, but at the same time ends up making the stereo imaging wider than anticipated. What's the correct approach?

I have a Carver tuner that offers something like that, and my Holman preamp provides fully adjustable separation. For FM multiplexing that affects stereo separation anyway it’s an interesting effect, but I don’t find myself using it. For digital sources like CDs, I have no trouble at all getting a gorgeous imaging effect using good speakers properly located.

Recordings are often a manipulation in that the instruments are recorded in mono and then panned between channels during mixing. With those, the effect is what it is. I wouldn’t want any such feature to be hardwired, though, because many good recordings use microphone arrangements that preserve staging. Even good vinyl LPs preserve separation well enough to spread the phantom image pretty well in my system.

Lots of speaker arrangements undermine stage imaging, and maybe such tricks are just a band-aid.

Rick “doesn’t like the hard separation tricks used by some FM radio stations” Denney

 

[Tim Link]

Recordings are often a manipulation in that the instruments are recorded in mono and then panned between channels during mixing.

Conceptually volume panning makes the most sense to me. The resulting sound waves cross the head at the correct angle for the location they are supposed to be coming from between the speakers. As an added benefit the signal will mix down to mono without creating nulls. The big problem is the cancelation nulls at the listener's ears created by the stereo pair of speakers playing sounds panned toward the center. So I'd still want some way to address that. It seems to me that crosstalk reduction should be applied aggressively to center panned images where the nulls are the worst, and not at all to signals panned hard left or right because in that case there are no crosstalk issues causing comb filtering.

 

[Multicore]

More on reproduction versus musical performance or interpretation.

I can take a composition, Ghosts, as I mentioned above is as good an example as any, in the standard language of music notation, e.g. as in #686, and

1. convert to MIDI
2. pay it using a VST instrument outputting a WAV audio file
3. use AI software to produce a score from the WAV

Sure there are some details but with a little care we can make this process idempotent and lossless. Each step here is reproduction, mechanical conversion from one format to another.

And we can use a well skilled human to convert the score to a WAV that step 3 can convert back to a score just as above but we would need an instrumentalist who can remove enough of the musical performance/interpretation that would be natural, customary, or expected in playing Ghosts for it to work as reliable reproduction.

 

[antcollinet]

Playing music is not the same as composing it. Each time somebody plays a composition they are reproducing it. Joni Mitchell commented on this. She said if someone paints a great piece of art, no one ever askes them, "Hey, paint me Starry Night again!" She liked painting because she only had to do a painting once. Now a copy of a painting made by a machine may have some constraints. Will it try to copy 3D effects, like thickness of paint in certain areas? Will it try to copy the canvas material properties so it feels the same? Will it try to scale to the exact same size, or blow it up bigger, or smaller? Will it use a reduced color set, maybe just black and white? Are these purely engineering decisions, or is it somewhat of a blend? Maybe there are engineering constraints that require some artful decisions to get a good looking result.

Another recent necromance of a 5 year old thread. It was all hashed over more than five years ago - why bring it up again.

And to answer your belated point, honestly - I am struggling to understand how an intelligent person can think that the process of performing a composition bears any resemblance to the process of reproducing a recording of that performance.

 

[Multicore]

Another recent necromance of a 5 year old thread. It was all hashed over more than five years ago - why bring it up again.

not meaning to speak for anyone else but maybe i wasn't here or didn't take part 5 years ago, maybe some topics aren't forever foreclosed because a internet forum conversation became quiescent, maybe some things are worth discussing again or one has new ideas.

And to answer your belated point, honestly - I am struggling to understand how an intelligent person can think that the process of performing a composition bears any resemblance to the process of reproducing a recording of that performance.

Take a look at #653. Axo1989 certainly counts as an intelligent person.

 

[Axo1989]

not meaning to speak for anyone else but maybe i wasn't here or didn't take part 5 years ago, maybe some topics aren't forever foreclosed because a internet forum conversation became quiescent, maybe some things are worth discussing again or one has new ideas.

I wasn't around here in 2020 either. We've had much discussion on reproducing live recordings vs actual live performance for sure, but not so much (that I recall anyway) on reproducing musical scores qua reproduction (or not). So I though the recent conversation here was sufficiently novel to get my interest.

Take a look at #653. Axo1989 certainly counts as an intelligent person.

Thank you, I hope so on a good day at least

Anyway I'm going to listen to Akchoté and think about your earlier post shortly, there's more to consider there ...

 

[Axo1989]

This whole recent thread qualifies as such.

You're welcome to think so of course, but I was thinking of the specific argument @RexrothPigeon made that loudspeakers don't have anything to do with 'reproducing' music files, but instead they do 'specific technical thing'. That poster hasn't elaborated since, however.

 

[antcollinet]

Axo1989 certainly counts as an intelligent person.

I think I acknowledged that - hence my confusion

 

[Axo1989]

From my experience as a musician who sometimes plays compositions and uses improvisation, as almost all musicians do outside the classical world, I can't really get behind this idea.

Recently I've been working towards making a recording of Ghosts by Albert Ayler. It is, without a shadow of doubt or possible controversy, the greatest tune to emerge in the 20th century and Don Cherry was right to insist that it should be the national anthem of the USA.

View attachment 506389

I can reproduce this, i.e. play it dead straight according to the score (a useful step in learning it) but that's an absurd thing to do. Afaict from the half dozen recordings of Ayler I've studied, he never reproduced it. What did Ayler and his bands do instead of reproduction? The verbs I find useful here are to perform and to interpret, both of which contrast with to reproduce. Since I'm playing it on guitar I also have to arrange it, a useful technical term.

Specifically for my recording, I want to imagine how Sonny Sharrock might have interpreted it, in a counterfactual history in which it would be on his album Guitar. Would Sonny Sharrock have reproduced Ghosts?

Noel Akchote has five different interpretations and arrangements of Ghosts in his collection last year Of Albert (Complete Plays Ayler, 2025). None of those seem like reproductions to me. What do you think of this delightful version, under 1-minute long?

Ghosts (Arr. Gary Windo, 1988), by Noël Akchoté

from the album Of Albert (Complete Plays Ayler, 2025)

noelakchote.bandcamp.com

N.A.D.'s 1989 Italian recording with Fred Frith on violin is high on my list too. Sharrock appears later on that album providing the solo to Bibbidi-Bobbidi-Boo (yes, that one).

This morning I had the idea to grow a manouch, don a fruity cravate, and do a Le Jazz Hot-style version of Ghosts.

Is it even possible to reproduce a composition like Ghosts? Yes but no.

I'll have a go at my response to this excellent post. That album not on Apple Music for my listening convenience so Bandcamp it is, and a nice purchase: 90 tracks! Four instances of Ghost that I could see (did I miss the fifth somehow?). And for background/comparison I tried two variations by Ayles from the album Spiritual Unity (the original album Ghosts is also missing from Apple's offering, so I haven't heard the basis).

I'm really profoundly ignorant of jazz history and listen to the broad genre not a lot (and I'm not even a fan of brass instruments generally). But (as we've discussed in a different thread) I enjoy avant-garde/experimental work in many other areas so not really surprising that I found Ayles' variations pretty wonderful. The part I liked less was (what I assume is) the original melody but that needs to be there for the improvisation/variation to make sense so no complaints. For the former reason (the melody) I can take or leave the version on Love Cry although the improvisations are still fun.

The Akchoté arrangements for guitar are no doubt more fascinating for someone knowledgable/familiar with that area of music but I enjoyed them no less. I expect I'll play the entire album through in the future more than once and enjoy it also.

Now to the question of the musical score. Obviously I agree that Ayles' variations push/break the boundaries of 'very similar to' qua reproduction. And while it would be possible to transcribe a less lossy score of a given improvisational performance, the reasons for (and efficacy of) doing so would be esoteric.

Firstly, while I certainly agree that certain levels of improvisation/interpretation will deviate sufficiently from an original score to be more accurately thought of as new works, but that isn't universally the case. It makes sense that you employ 'perform' and 'interpret' in the former cases and generally for this type of music. But you seem to use 'reproduce' and 'perform' as mutually exclusive antonyms rather than the synonyms that they are?

Secondly, looking at the Akchoté album materials—he calls the collection Transcriptions which triggers my word-associating brain, but I digress, and I know the technical meaning is straightforward—he proves score for his variations. This one for the shortest instance (and your listening recommendation) at track 89:

I don't read music or play an instrument unfortunately, but I'd guess the performed/recorded piece is more likely to reproduce this score than the example in your post? Whether you (or anyone/everyone else) considers the score too lossy or otherwise lacking to be 'reproduced' is an open question.

To digress now, the 90-track length from Akchoté reminds me of a different series of variations. While Apple Music let me down on some searching here, it's generally good for me. My 'replay: all time' list that appeared recently is topped by Arca, likely due to repeated listening to their release Riquiqui;Bronze-Instances(1-100) being variations on the original track Riquiqui using the ML tool Bronze. I don't have details of the workflow but assume the original track was training corpus for generative output. So the track becomes the score (plus any prompts employed) and the machine performs the variations/improvisations—although I like the term 'instances' (playing the link needs an account unfortunately but it's illustrative anyway):

https://soundcloud.com/arca1000000%2Fsets%2Friquiqui-bronze-instances-1

And of course (but pushing the boundary a bit also) I'd say these instances reproduce the original track. But not in the 'play it dead straight according to the score' sense that you used the term.

 

[Axo1989]

... And to answer your belated point, honestly - I am struggling to understand how an intelligent person can think that the process of performing a composition bears any resemblance to the process of reproducing a recording of that performance.

The parallels seem obvious/unavoidable to me. Your 'any resemblence' is surely rhetorical?

A CD contains a digital recording of music, just like sheet music does. Sheet music uses a lossy encoding, while the CD is a lot more precise. And while a DAC converts music from PCM on a CD, an orchestra converts it from musical notation on a piece of paper.

A CD is similar to sheet music, while an orchestra is similar to a DAC + speakers One could also note some similarity between a DAC clock and the conductor

Assume @pkane is playing it semi-straight and not totally sending me up (a possibility) this post would be a more concise version of my line of thought. Do you disagree, or really not see any resemblance at all?

 

[antcollinet]

'any resemblence' is surely rhetorical?

Not at all rhetorical - which you should understand (but seem not to) when I point out I used the words "process of"

 

[Multicore]

I'll have a go at my response to this excellent post. That album not on Apple Music for my listening convenience so Bandcamp it is, and a nice purchase: 90 tracks! Four instances of Ghost that I could see (did I miss the fifth somehow?). And for background/comparison I tried two variations by Ayles from the album Spiritual Unity (the original album Ghosts is also missing from Apple's offering, so I haven't heard the basis).

I'm really profoundly ignorant of jazz history and listen to the broad genre not a lot (and I'm not even a fan of brass instruments generally). But (as we've discussed in a different thread) I enjoy avant-garde/experimental work in many other areas so not really surprising that I found Ayles' variations pretty wonderful. The part I liked less was (what I assume is) the original melody but that needs to be there for the improvisation/variation to make sense so no complaints. For the former reason (the melody) I can take or leave the version on Love Cry although the improvisations are still fun.

The Akchoté arrangements for guitar are no doubt more fascinating for someone knowledgable/familiar with that area of music but I enjoyed them no less. I expect I'll play the entire album through in the future more than once and enjoy it also.

Now to the question of the musical score. Obviously I agree that Ayles' variations push/break the boundaries of 'very similar to' qua reproduction. And while it would be possible to transcribe a less lossy score of a given improvisational performance, the reasons for (and efficacy of) doing so would be esoteric.

Firstly, while I certainly agree that certain levels of improvisation/interpretation will deviate sufficiently from an original score to be more accurately thought of as new works, but that isn't universally the case. It makes sense that you employ 'perform' and 'interpret' in the former cases and generally for this type of music. But you seem to use 'reproduce' and 'perform' as mutually exclusive antonyms rather than the synonyms that they are?

Secondly, looking at the Akchoté album materials—he calls the collection Transcriptions which triggers my word-associating brain, but I digress, and I know the technical meaning is straightforward—he proves score for his variations. This one for the shortest instance (and your listening recommendation) at track 89:

View attachment 506978

I don't read music or play an instrument unfortunately, but I'd guess the performed/recorded piece is more likely to reproduce this score than the example in your post? Whether you (or anyone/everyone else) considers the score too lossy or otherwise lacking to be 'reproduced' is an open question.

To digress now, the 90-track length from Akchoté reminds me of a different series of variations. While Apple Music let me down on some searching here, it's generally good for me. My 'replay: all time' list that appeared recently is topped by Arca, likely due to repeated listening to their release Riquiqui;Bronze-Instances(1-100) being variations on the original track Riquiqui using the ML tool Bronze. I don't have details of the workflow but assume the original track was training corpus for generative output. So the track becomes the score (plus any prompts employed) and the machine performs the variations/improvisations—although I like the term 'instances' (playing the link needs an account unfortunately but it's illustrative anyway):

https://soundcloud.com/arca1000000%2Fsets%2Friquiqui-bronze-instances-1

And of course (but pushing the boundary a bit also) I'd say these instances reproduce the original track. But not in the 'play it dead straight according to the score' sense that you used the term.

Thank you for that. There are a few things to consider.

First the semantic. As I suggested in #690, performance can and sometimes does encompass reproduction. The word reproduction, in my view, implies some kind of fidelity. An old Xerox photocopier can reproduce a text document with perfect fidelity, as can a 19c. copyist with a quill pen, or a fax machine since fidelity is in getting the text right. These systems cannot adequately reproduce a color photo or oil painting. We would need to upgrade such systems so that the output represents the input to a level of fidelity that we accept as a reproduction.

Performance does not imply fidelity. Performance can use fidelity but doesn't always need to because interpretation may be permitted, more or less, depending on the specific traditions and context involved. I used the example of Ghosts because its tradition licenses broad interpretation and as much as demands personal interpretation from performers. What music might demand the minimum? Paganini, perhaps?

So performance is one thing and reproduction is another thing and one doesn't imply but can involve the other and visa versa and they may overlap.

I bang on about this because the wiggle room is very important to me. As a musician and in the audience, the magic is in the infidelity, i.e. what we do in the wiggle room. To me this seems obvious but I suppose it might not be to others and especially on a forum that's most concerned with playback of recordings. I have a complicated relationship with recordings (see 3. here).

---

The Gary Windo arrangement goes a bit beyond arrangement and is perhaps a derived composition. It forces the A section and first half of the B section (as shown in the chart I provided) into a strict folk rhythm, retaining as much of the melody as can be fit into that, and discarding the intro and second half of B. The tradition implied in Windo's version has much stricter demands (less wiggle room) that the the successful performer is likely to respect.

Akchoté is usually relentlessly creative but in that recording of Windo's version his contribution is modest and he's respectful. He does the boom-chick rhythm part conventionally using the chord symbols (C, F, G7 etc) and the melody has some ornaments and distinctive staccato but that's about it. It's not a perfectly straight reproduction of the score but it's close, more so than most people play Ayler's Ghosts.

---

I found Riquiqui;Bronze-Instances on Qobuz and listened to a few and I agree that instances is a good word here. Your guess about how it was done seems likely. I'm not sure what to say about it with respect to the current discussion because, at the risk of sounding like some weirdo, my concern is with the capacity for free communication from and to the unconscious mind, communion of the ineffable, or a spiritual unity, if you like. Do you remember the thread Recordings as Simulacra? @Curvature nailed it in this post:

Recordings as simulacra

Borges in turn may have drawn from Lewis Carol's 1889 story Sylvie and Bruno Concluded which also featured a 1:1 scale map I thought Swift had done it earlier, but maybe imagined that) Thanks—I wasn't aware of that reference.

www.audiosciencereview.com

How AI tools might fit in to all that, I have, thus far, no idea.

 

[Jim Taylor]

If a composer gives a musician a composition and the musician plays the composition, is he performing, translating or reproducing?
If an audio engineer makes a recording of that performance, is that recording a reproduction or a copy?
If a company makes 1,000 files of the recording to sell, are they copies or reproductions?
If a musician listens to one of the files and plays the music exactly like the recording (so that no one can tell the difference), is he copying, performing or reproducing?
If a publisher publishes the composition in sheet music form, are they reproducing it?
If they are, what about the musician who copies the performance so well that no one can tell the difference? Is he copying the performance or performing the composition?
If the original composition is performed by a musician and then played on a speaker, is the original musician reproducing or the speaker reproducing?
And what are they reproducing? Is it the composition, or are they reproducing a reproduction?

Just asking for a friend.