Help needed from experienced session players in or around London

Help needed from experienced session players in or around London

Hi all,

You might remember me as the one posting about a year ago on "computer-generated session tunes." We really enjoyed the variety of responses you had, and now wish to see who is interesting in helping us in more specific directions.

In short, we wish to meet with experienced session music players to discuss some of the 3000 tunes in this collection: http://www.eecs.qmul.ac.uk/~sturm/research/RNNIrishTrad/folkrnn/folkrnn_vol01of10.pdf
(It is about 35 megabytes, so don’t download on your phone!) The second page describes what we are interested in.

Please contact me if you want to contribute, and live in or around London.

Many thanks.
-Bob.

Re: Help needed from experienced session players in or around London

Interesting project. Best of luck with it. I took a peak and was surprised to see incomplete (or what appears to me to be incomplete) notes in a bar - e.g. Bar 6, Tune 3 - thinking that that would be avoided throug the coding algorithm.

Am I mssing something about the nature of the project?

Regards,
Brian

Re: Help needed from experienced session players in or around London

The PDF is indeed a huge motherflipper of a file as Bob has said. If you want the preview of what the people involved in this project are looking for, here is the content of page 2:


"The 3,000 tunes in this session volume have been generated entirely by a computerised system (named folk-rnn) trained on 23,636 tunes contributed by users of thesession.org. Each tune appears as ABC and in staff notation, and is hyperlinked to a synthesised realisation. As the developers of this system, we want to know what it has learned to do. What has it got right, and what has it got wrong? Since such information can greatly help us to improve the system, and adapt it for a variety of di erent uses, we are seeking input from people experienced in playing session music. Examples include:

1. How hard is it for you to find tunes in this volume that you think are close to the kind of
music you encounter at a session? If you can, identify some tunes that fit this description, and
explain why.

2. How hard is it for you to find tunes in this volume that you think are far from the kind of
music you encounter at a session? If you can, identify some tunes that fit this description, and
explain why.

3. Pick a tune in this volume that you think is close to the kind of music you encounter at a
session and say how it could be improved, if at all.

4. Can you find a tune in this volume that is close to one that already exists?

Other questions are possible as well. If you are interested in contributing, please send an e-mail to
the project leader, Bob L. Sturm "

Re: Help needed from experienced session players in or around London

A potential problem that I could see with training the machine on 23,636 tunes here is that there’s a lot of newly-composed dross and non-Irish tunes in the database at this stage. I’m only speculating now, but I imagine the real pure-drop Irish tunes are probably outnumbered by tunes that arguably are not part of the Irish tradition at all.

It might be useful to select a subset of the available tunes that are verified by human musicians as being either traditional Irish or at least ‘sounding traditionally Irish’ and use those as a baseline for training the machine. That’s just my two cents :-)

Re: Help needed from experienced session players in or around London

Thank you so far for the comments.

Brianmo: The system has just learned to produce ABC and sometimes appears to miscount. (Its ability to count is really compromised when it is seeded with something outside the conventions it has seen.) That is certainly one aspect needed for its improvement.

Colman O’B: Thanks for posting the text! (I should have thought about that. :) Our system is trained on ABC collected here at thesession.org, and so we do not mean to say that it should be creating Irish tunes. In the end, what we are interested in is creating an assistive creative system for a composer or student or player or … And not necessarily in the session vein. But we are starting here because I like to play this kind of music, and you all have done a fantastic job collecting the data and making it accessible. :)

Re: Help needed from experienced session players in or around London

Hey Bob,

I’m interested - PM me?

Re: Help needed from experienced session players in or around London

I am pretty excited that our robot overlords will have been steeped in ITM.

But seriously, if you’re saying this was created by a program that just studied a few thousand tunes and then created these files, I’m impressed. I just skimmed, but they seem playable. Not always friendly for the whistle register or fifth-tuned instruments, but I could detect clear echoes of what I hear a lot in trad tunes.

Thanks for sharing! Tune 87 I liked. Sounds familiar.

Re: Help needed from experienced session players in or around London

To be honest, I haven’t much interest in this, but I just scrolled through the first 50 tunes listed out of idle curiosity. The commonest key seems to be G major, but not a single one is in D major, which appears to me to be very odd. None of the 50 were in B minor, E minor / dorian, C major, A major or A mixolydian. Now this may be "corrected" if you want to put it that way, throughout the rest of the tunes - I’m not trawling through them all to find out - but how would you explain that ?
I had a look at the first 3 tunes, and listened to the 3rd. The first 2 are garbage - there is none of the "question / answer" type of repetition within either tune which is the hallmark of traditional Irish dance music, a topic discussed here fairly recently. The 3rd one sounded a bit more convincing, but there are irregularities in a few of the bars, as "Brianmo" mentions above.
Sorry, but that’s as much time as I’m prepared to commit. I hope others can be of more help to you.
Best of luck, Kenny.

Posted by .

Re: Help needed from experienced session players in or around London

Hi Bob,
What was the nature of the Session.org submissions?

Where did the material come from?

Posted by .

Re: Help needed from experienced session players in or around London

I just had a first attempt at skimming through a bunch of the tunes and trying them on the whistle, and my overall feeling from the random tunes I picked is that if you go bar-by-bar there are lots of very ‘traditional sounding’ phrases, but as Kenny says they tend not to actually go together to form coherent tunes as a whole.

Tune 39 was actually sounding like quite a plausible traditional tune up until it ended on a low E when it clearly should have ended on a G instead :-)

Re: Help needed from experienced session players in or around London

Hi all,

Tøm: message sent!

Alex Henry: Thanks! Tune 87 is a nice one, but can certainly be improved by adding repeats after the tune and turn. I like how it switches between D minor and C major.

Kenny: Thanks for looking through it at least. Sharp eye on the appearance of modes. Our system only generates tunes in four modes with root C: Cmaj, Cmix, Cdor, and Cmin. (We transposed all session transcriptions to be in those four modes to reduce the dimensionality of the learning problem.) We then transposed the generated output to Ddor (if in Cdor), Gmaj (if in Cmaj), Dmix (if in Cmix), and Amin (if in Cmin). We made this decision based on a conversation with a local session group (who only play sharps). It could be done in a smarter way than this, for sure. Tune 1 is garbage. Can tune 2 be improved by adding repeats at the end of the tune and turn? Tune 3 definitely has counting issues.

Ergo: The data comes from a datadump. All 13 MB are here (in a cleaned version): https://github.com/IraKorshunova/folk-rnn/blob/master/data/sessions_data_clean.txt

Colman O’B: Thanks for trying some out! Tune 39 is interesting. It has a pickup, but this is not accounted for until the end of the entire tune. I agree about the ending. This is something to tune in the system — though maybe the computer means for one to play the tune once more after the turn. :) Tune 40 is a royal mess. Tune 101 is really fun.

Re: Help needed from experienced session players in or around London

"Ergo: The data comes from a datadump."

Of what? What were the contributions of S.Org members?

Posted by .

Re: Help needed from experienced session players in or around London

Ergo: We used the entire collection of tunes on thesession.org (as of June 2015). These are made once a week: https://github.com/adactio/TheSession-data We did put in a significant effort cleaning it up, however, as many entries were alternative endings, or comments, or jokes … some jokesters have contributed ABC transcriptions of John Cage’s 4’33"!

Re: Help needed from experienced session players in or around London

Had a quick look again while the flute’s out. I tried #6, but in D major, and not Dmix, and it almost works, although it opens very much like "The Floating Crowbar". Could work with a bit of "tweaking". Possibly comes under your categories 3 and 4.

Tune #4 - the jig - spends far more time up on "high c" than any other Irish tune I’ve ever come across.

I’ll keep looking when I find the odd moment, Bob, and let you know if I come up with anything.

Posted by .

Re: Help needed from experienced session players in or around London

"Ergo: We used the entire collection of tunes on thesession.org (as of June 2015)."

And you transformed them into new tunes?

Posted by .

Re: Help needed from experienced session players in or around London

From skimming the PDF, I assume you’re using some kind of Markov approach? That’s certainly what it feels like, with each phrase responding to the previous one but not to any kind of higher structure.

What would be interesting would be some sort of branching procedure that understands structure between parts, phrases, bars and phraselets.

Posted by .

Re: Help needed from experienced session players in or around London

Kenny: Thanks! I see the resemblance between the tune of #6 and The Floating Crowbar (https://thesession.org/tunes/457). The system seems to have forgotten to put in the repeats after the tune and turn. Do you think transposing #4 down to G makes it more plausible?

Ergo: The system has learned to model the ABC language from the transcriptions thesession.org users have contributed. It is a simple matter to turn the system around and have it generate new transcriptions, many of which bear some resemblance to the training material. So, in a very correct sense, the system is transforming real tunes into new tunes via low-level principles it has learned from the collection.

Calum: The system can be seen as a Markov chain, but one with a lot of memory and high complexity. More formally, it is a deep recurrent network consisting of three hidden layers of 512 LSTM units each. It models tunes one token at a time in a sequential manner — which is only one approach to music composition. When you say, "higher structure", what do you mean?

Re: Help needed from experienced session players in or around London

That’s going beyond my pay grade! I mean simply things like in a two parted tune, the second part will usually be in a higher register, for example. My very vague idea is that a more idiomatic composition system might have a "structure" generator, which understands the distribution of this "high level" structure, and then feeds it to lower level generators that then populate that structure with phrases (which might then be fed to another generator to produce the actual melody).

The obvious question is what is that high level structure and how is it represented. Don’t know. Some sort of dimensional analysis, perhaps (PCA etc)?

Posted by .

Re: Help needed from experienced session players in or around London

Calum: Ha ha. We train the system only on the ABC code at thesession.org, and give it no other information about what the symbols mean, how many of one kind should appear before another kind, structure, conventions, etc. Among the 30,000 new transcriptions it has generated (the first 10% of which are in Vol. 1), we see many that have an AABB structure, correct counting, rhythmic consistency, a turn in a higher register than the tune, acceptable V-I cadences, etc. Examples are not hard to find: tunes 8, 9, 16, 31, 39, … 1772, 2231, 2721, 2722, … Of course, among these there are many odd ones too, and some that are just downright implausible.

Re: Help needed from experienced session players in or around London

Any attempt at a system to assess if tunes are any good or not? To save humans having to listen through 30,000 tunes?

Is this part of something like that?

Re: Help needed from experienced session players in or around London

Well……..
…..you guys turned up at a session I sometimes go to a few weeks ago, and obviously interested one or two people there.
Personally I don’t see the point.
This is ultimately going to be the subject for your thesis or dissertation, right ?
So far computers have proved they can play chess, but they can’t make people believe that they are human.
Also, they don’t seem able to write interesting playable tunes, at least with the algorithms that you have written.
That’s about it, as far as I can see.
I do hope that you give full credit to everyone who has assisted you in this, what I would call pointless, endeavour.

Re: Help needed from experienced session players in or around London

I can see three stages of development here:

1. A machine learning algorithm to enjoy tunes. (this part)
2. A machine learning algorithm to tell us which of the first tunes are any good.
3. A machine learning algorithm to listen to the resulting tunes and enjoy them without us.

Then we can give up completely and let the machines take over :-)

Posted by .

Re: Help needed from experienced session players in or around London

Bob - thanks for your explanation.

Guernsey, I’m sort of with you on this. Not only is there no point - sorry - but some 3000 machine-generated tunes are now in the public domain, which troubles me.

Posted by .

Re: Help needed from experienced session players in or around London

I wouldn’t worry Ergo. They seem to be less accessible than the dross amongst the good transcriptions here.

I would be far more interested in what came out of this sort of thing if what went in was a coherent set of transcriptions such as, say, those of O’Neill’s 1850. Maybe that’s already been done and so is of little academic interest.

Re: Help needed from experienced session players in or around London

Thanks all!

David50 and pizak: Indeed, that is where we are heading: an automated critic that can provide feedback during training. (This is related to a hot topic right now in machine learning called adversarial training. Essentially the critic system will take an output generated by the other system and say whether it thinks it is real or fake. The goal of the system is to learn how to fool the critic system.)

Guernsey Pete: That’s great you were in the room when we came by! (Once I buy a D/G melodeon I will join that session.) I appreciate your scepticism. We are not trying to build a system that will fool people into thinking it is human, or trying to equal human composition. I do think that is pointless! Instead we are aiming to build an automated assistant for music creation, education, … . We also aim to adapt it to other stylistic conventions of music. In my experience with using the system, it certainly doesn’t produce great stuff all the time. Sometimes, however, there are some really interesting results. Here is an interesting and playable tune produced _entirely_ by the system: https://www.youtube.com/watch?v=YMbWwU2JdLg . And here is a tune created with a few edits of the system output https://www.youtube.com/watch?v=Wm5yQR336cI . (Please forgive my horrible playing! :) In a completely different direction, here is a march composed in collaboration with the system: https://soundcloud.com/sturmen-1/the-march-of-deep-learning . As for credit, our recent publication (https://drive.google.com/open?id=0B1OooSxEtl0FcTBiOGdvSTBmWnc) acknowledges Keith and the many contributors of thesession.org! :)


Ergo: That raises an interesting point: does the existence of the 30,000 tunes generated by our system (available in volumes 1-10) affect the value of the real tunes at thesession.org? Or maybe the troublesome feeling is something else. Quality?

David50: A common technique used in deep learning to build language models is to train the network first on a bunch of stuff (like millions of wikipedia entries). From this the system learns fundamentals about sentence structure. Then the system is tuned on a smaller collection, e.g., poetry, to specialise it. Humans kind of do the same thing. One learns how to listen to all the sounds around us, and then build upon this to develop musicality, understand speech, and so on. So, a next step could be to take our system and specialise it on Dow’s 60 or O’Neill’s 1850.

Re: Help needed from experienced session players in or around London

"Ergo: That raises an interesting point: does the existence of the 30,000 tunes generated by our system (available in volumes 1-10) affect the value of the real tunes at thesession.org? Or maybe the troublesome feeling is something else. Quality?"

My concern is that some people, somewhere and sometime, may consider one or more of these tunes - maybe all of them? - to be actual traditional tunes. That’s my concern.

Posted by .

Re: Help needed from experienced session players in or around London

I just opened up the .pdf and paged down to a random tune (actually #4) and I like it :)

If this is the result of an algorithm, it’s doing pretty well. That’s just one tune, though. I’d have to choose many random tunes there and gauge the result, to avoid confirmation bias, I guess.

I like your idea, Bob Strum.

Re: Help needed from experienced session players in or around London

@Bob Sturm. Wikipedia entries conform to a subset of grammatical conventions and the vast majority have been edited to weed out departures from those conventions. Not so the tune database here.

Your system does not seem to have learned the analogue of ‘the fundamentals of sentence structure’ - as pointed out by Brianmo above it doesn’t always match the notes in the bar to the metre. What chance that is due to training it on stuff with typos?

Re: Help needed from experienced session players in or around London

Ergo: I hope that I have made those things clear in the introduction! Maybe we can put a watermark on each page saying, "FAKE". :)

Jim Dorans: Thanks for trying it out. How come you think #4 is good? How could you improve it (other than transposing it way down :)

David50: Nope, it doesn’t always count right (but it actually does a lot of the time). One contribution to this problem is that the system has not learned on enough data. With only 23,000 transcriptions and over 5 million parameters to learn, there is definitely going to be some underfitting. Also, the training data is sure to have problems, as you point out. I have found transcriptions in thesession.org in which people don’t put in measure bars! I have tried to weed as many of these out as possible when I cleaned the data.

Re: Help needed from experienced session players in or around London

[*Jim Dorans: Thanks for trying it out. How come you think #4 is good? How could you improve it (other than transposing it way down :)*]

Bob - I think it’s good for its ‘austere’ melody.

I wouldn’t bother transposing it - it works fine on fiddle. The high C is quite common in tunes - that said, the high C in bar#7 does sound a little bit odd - maybe drop that one down an octave?

The other high Cs are fine, and they ‘go with the flow’.

The bar arrangement on the page is a bit odd - should really be 4 lines of 4 bars per line, but I’m sure that was picked up from the source, and not done by the software.

Re: Help needed from experienced session players in or around London

"Technology does only one thing- it tends toward efficiency. It has no aesthetics. It has no ethics. Its code is binary.

But everything interesting in life- everything that makes life worth living- happens between the binary. Mercy is not binary. Love is not binary. Music and art are not binary. You and I are not binary." ~ T Bone Burnett

Posted .

Re: Help needed from experienced session players in or around London

Agree, irishfiddle.

Computers have a way of finding work regardless of whether or not there’s an actual "need." They can do a lot of good, but they can also be a waste of time.

Posted by .

Re: Help needed from experienced session players in or around London

"Computers have a way of finding work regardless of whether or not there’s an actual "need." They can do a lot of good, but they can also be a waste of time."

That’s an odd way of looking at things, along with your apparent fear of machine-made tunes escaping into the wild :-D

The tradpocalypse is nigh!!! Repent!! (it is permitted to make use of a stool while repenting).

I’m not sure computers really "find work" at all. This project appears to be a purely academic endeavor in the area where computer science meets music theory. The whole idea of algorithmic music composition can be pretty fascinating to some people who have a foot in each of those two worlds, and there are certainly people who read these discussions who fall into that category.

Traditional Irish dance music has a limited set of structures in terms of particular rhythmic forms, and particular keys/modes, so on the surface it appears to be a pretty good candidate for representation within computer systems, although as one can see from the project’s set of output tunes it might be surprisingly complicated to get right.

Surely there’s merit in studying such things for the sake of challenging your own abilities or increasing the body of knowledge in the area. I don’t think it’s a whole lot different from a herpetologist searching for an undiscovered species of frog in the amazon or a linguist trying to reconstruct Proto-Indo-European. It’s a niche interest within a niche interest that will go under the radar of all but those most interested in the area and won’t have the slightest effect on the lives of the vast majority of us. It’s probably less useful in practical terms than building algorithms to direct robot vacuum cleaners or self-driving cars, but why knock it as a field of study? Most career academics that I know are deeply buried in some very esoteric trench in their field.

I don’t think Bob and his team are aiming to replace the whole repertoire for a machine-generated alternative, replace your session mates with robots, or program your fiddle to play itself.

Re: Help needed from experienced session players in or around London

Thanks for these very good comments!

When it comes to some academic research, like spending billions of euros to find a single subatomic particle that may not even exist, or studying sexual preferences of insects, it can be hard to see the point, how this work impacts society, and why it isn’t just a big waste of money. However, answers to these fundamental questions can have enormous impacts that could never have been envisioned, e.g., breeding out disease carriers, uncovering new sources of energy, etc. One of my favorite examples is the X-ray, which came out of basic research studying things _that people couldn’t even see_. What’s the use? Nowadays the x-ray machine is an indispensible tool for medicine. (For a while too, x-ray machines appeared in shoe stores so one could see how well a pair of shoes fit their feet!) There was also research in radioactivity. What possible use could there be for matter that is in constant decay? That radiation is now a powerful way to treat cancer, and power space probes.

I am not working in basic science, but instead engineering. So, what’s the use of what we are doing in this project? Consider a piece of engineering that many people now take for granted: automatic speech recognition. When that research developed in the 1950s, engineers were trying to get a "computer" (something that would take up an entire floor of a building) to detect when a person spoke a single digit, "one", "two", etc. That surely seemed silly! Why would anyone want to speak to a "computer"??? With further research, computers were made smart enough to detect connected digits; then simple words, then sentences, and so on. Overnight, this technology in the 1980s saved the US telephone company AT&T an enormous amount of money because what once was done with manual switchboards was replaced by a menu driven speech recognition system. Sure, we all get annoyed when this technology doesn’t work; but nowadays, speech recognition technology has advanced to such a degree that we all have it in our little cell phones, i.e., Siri, Google, etc. This completely natural way of interacting with our computing devices could never have been foreseen in the 1950s. We even have automated the captioning of videos on YouTube! (Imagine if we had to caption all videos manually!)

Coming back to our project, what’s the point? Colman is right: we are not trying to replace musicians, or replicate a tradition. We are trying to get a computer to arrive to some level of understanding of a product of human culture. Why? One reason is that the cultural sphere remains a exciting frontier for integrating computer technology. Most people probably don’t know the true extent to which computers and software are already integrated with the production of culture — the entire pipeline is digital. But we have so far to go, beyond the Spotifys and Netflixs. What’s the use? One possible use of this can be in a music transcription scenario: you are at a session, you want to see the dots of the group playing, so the music transcription system listens and works with the knowledge we have given it of the tradition to produce the dots, the name of the tune, its history and variations, etc. It needn’t be session music, but that is where we are starting for many of the reasons Colman states … and you friendly practitioners have provided an excellent dataset!

I hope this makes some things much more clear and exciting for you!

Re: Help needed from experienced session players in or around London

"That’s an odd way of looking at things, along with your apparent fear of machine-made tunes escaping into the wild :-D:

OK. For the sake of argument, let’s take our own Jim Dorans here, who likes the tune #4.

What’s to keep him from going out and playing that tune in sessions? The tune, under a name, perhaps, given to it by Jim, would then be "in the wild." Yes?

Posted by .

Re: Help needed from experienced session players in or around London

Ergo: "What’s to keep him from going out and playing that tune in sessions? The tune, under a name, perhaps, given to it by Jim, would then be "in the wild." Yes?"

I agree, there’s nothing to stop him. I just don’t see the problem with it getting out.

If the tune is shyte then nobody will bother to learn it and it will disappear quickly and quietly.

If the tune is a good one, and someone does a nice job of playing it in a session or on a recording, then maybe it will catch on and a few people will learn it. Maybe it could even creep into the ‘standard session repertoire’.

And so what? What would be the problem with that? If the tune is good, does it really matter whether it’s of unknown origin, composed by Tommy Peoples, or the product of a form of (human-built) artificial intelligence exposed to the conventions of traditional Irish music?

Re: Help needed from experienced session players in or around London

@ Coleman. But it has been "exposed to the conventions of traditional Irish music" mixed up with other stuff.

What sense of the way Irish music works I get from the database relies in part on me being told which tunes are not Irish. I thought the nicest aspect of the tune in one of Bob’s youtube links were the Scotch snaps.

Re: Help needed from experienced session players in or around London

"If the tune is good, does it really matter whether it’s of unknown origin, composed by Tommy Peoples, or the product of a form of (human-built) artificial intelligence exposed to the conventions of traditional Irish music?"

Yes, it matters tremendously.

Note to Bob on Siri: do you know anyone who uses it? I don’t.

Posted by .

Re: Help needed from experienced session players in or around London

"Yes, it matters tremendously"

Why?

I know your comment re: Siri was directed towards Bob, but I can tell you plenty of people use it at least sometimes. Not me, because I can’t stand anything Apple, but lots of people I know definitely use it.

Re: Help needed from experienced session players in or around London

It would be very interesting to find out if there was a significant number of "acceptable" tunes, according to the style and/or individual musicians’ preferences. What went wrong with the unacceptable tunes, and what went right with the others…

Re: Help needed from experienced session players in or around London

What is ARTificial, the computer composing tunes or the human who has a musical style that is a replica of someone else’s style and who couldn’t compose a ditty?

Hmmmm, logical enough to me

Re: Help needed from experienced session players in or around London

It is not unexpected to see that the provenance of a tune is essential to some, while for others it is not. Any practice of music has conditions or requirements of authenticity. I don’t think Jim Dorans is a walking "dirty bomb" when it comes to session music, but that is neither here nor there now that we have named tune #4, "Jim Doran’s." :)

Re: Help needed from experienced session players in or around London

Ergo: "Yes, it matters tremendously"

I too would be interested to know why you think it matters tremendously. (I’m not saying it doesn’t). Is it because of some sort of artistic ethic? Is it because a tune composed by a machine is, de facto, artistically inferior?

Most tunes that we (traditional musicians, actual and aspiring) play, whether or not they have a known composer, have been shaped by the hands of many musicians - they have had the spiky bits rounded, the difficult bits made easier, the fussy bits made simpler, the boring bits made more interesting, the clunky bits made smoother and so on. If a robot-generated tune were successfully introduced into the session repertoire it would inevitably undergo the same process at the hands of us humans, ergo it ends up being (in part, at least) a human creation, anyway.

Re: Help needed from experienced session players in or around London

I think there is genuine fear in expecting sessioneers to play "long march of deep learning". If it did show up. some notes seemed particularly out of context

Re: Help needed from experienced session players in or around London

Well, I reckon it should be called "Bob Sturm’s No 4".

I like the tune.

Couldn’t be simpler :)

The laws of probability alone would suggest that there are many more good tunes in that big file. I for one, will go through them all in my own time.

Re: Help needed from experienced session players in or around London

I secondly second Jims reckoning. The reel caught my attention. For my two kernels worth

Re: Help needed from experienced session players in or around London

"I too would be interested to know why you think it matters tremendously."

I like to think that music comes from somewhere that’s living. Otherwise it’s…something else.

Algorithms, computers - they’ll always be predictable. Not so much humans.

Posted by .

Re: Help needed from experienced session players in or around London

"I know your comment re: Siri was directed towards Bob, but I can tell you plenty of people use it at least sometimes."

I’m curious - specifically what for?

Posted by .

Re: Help needed from experienced session players in or around London

Thanks for posting this, Bob.
I have no doubt you’ll find some experienced session players to bring out the best from your collection of tunes.

Posted by .

Re: Help needed from experienced session players in or around London

Jim Dorans: Thanks! Let me know, maybe by personal email (b.sturm@qmul.ac.uk) if you want to meet and discuss in more depth.

Ergo: No doubt, music is a human activity. I prefer to see our little system here as producing only ABC transcriptions, not music. When humans get involved, that is when music can happen. Regarding Siri, and other speech recognition applications on portables, many use them because it is a far less clunky way to get and send information than typing letters on a little screen. Such applications also make devices accessible for people who can’t see well enough or have the dexterity to type. There are many reasons why Apple and Google and research groups around the world devote a lot of time and money to improving this technology that once was seen as ridiculous. :)

Re: Help needed from experienced session players in or around London

Ergo: I should also mention that speech recognition provides much more than just a fun means of controlling devices. It is also being applied to the automatic translation between spoken languages. I have seen it used to enhance communication between Chinese and English people. The science fiction of the "universal translator" in Star Trek is coming to life. No one can really deny the utility of such a technology. :)

Re: Help needed from experienced session players in or around London

"No one can really deny the utility of such a technology."

"Regarding Siri, and other speech recognition applications on portables, many use them because it is a far less clunky way to get and send information than typing letters on a little screen. Such applications also make devices accessible for people who can’t see well enough or have the dexterity to type.’

I don’t want to pull this off track any more than I already have. And I get that voice recognition can be useful. But if anyone has a second, I’m still curious what real people are using Siri for. Also, the wonders of technology aside, I think it’s reckless to send 3,000 machine-created fiddle tunes into the world.

Posted by .

Re: Help needed from experienced session players in or around London

[*Jim Dorans: Thanks! Let me know, maybe by personal email (b.sturm@qmul.ac.uk) if you want to meet and discuss in more depth.*]

@Reckless Bob - for sure I’ll be reporting back, after I’ve been through some more of your tunes!

Re: Help needed from experienced session players in or around London

Fred Frith would love that one!

Re: Help needed from experienced session players in or around London

@Jim Dorans, I spent some time with your #4. What do you think of these changes? (The turn was quite problematic.)

M:6/8
K:Ddor
dfa agf|e2c cfe|dfa agf|e_ba gfe|
dfa agf|e2c cBc|def ge^c|gag f2e:|
def agf|e2g _bag|fga c’_ba|g2_b d’c’_b|
a_bg f2a|gaf e2g|f2g d2^c|ea2 gfe:|

Re: Help needed from experienced session players in or around London

Ergo: "I like to think that music comes from somewhere that’s living. Otherwise it’s…something else."

I think most of us would agree with that. But it could be argued that a computer-generated tune is not music until somebody plays it, or ‘puts music into it’. The question is whether, at some time in the future, people find it preferable to listen to music composed by *and played by* machines over music played by humans. There have been machines playing music since the 18th Century - musical boxes, player pianos, steam organs and barrel organs. Each of these still has its curiosity value but, in a wider musical context, they have largely been replaced by machines giving ever more faithful reproductions of actual human beings playing and singing. I would not pretend to be able to predict the future of music and its relationship with technology, but trends suggest that there is a recurrent tendency for people to be drawn to a human element in music, whether that be its composition, interpretation or execution.

Re: Help needed from experienced session players in or around London

[*@Jim Dorans, I spent some time with your #4. What do you think of these changes? (The turn was quite problematic.)*]

@Bob, I like them. Although the original was fine for me, as it was, your changes have ironed out a bit of the angularity, and possibly made it more accessible to the average player. I’d keep both, and use the changed part as the variation, I reckon.

Re: Help needed from experienced session players in or around London

PS - I hope you, the computer and Jim Dorans come up with a nice version of tune #4 :)

Posted by .

Re: Help needed from experienced session players in or around London

Ergo, why do I think you’re just being less than sincere?

Posted by .

Re: Help needed from experienced session players in or around London

[*PS - I hope you, the computer and Jim Dorans come up with a nice version of tune #4 *]

Sorry to disappoint you, but we already have :)

Re: Help needed from experienced session players in or around London

@Ergo: I don’t think you are being insincere, just genuine in your skepticism. :) I am very interested to know how #4 can be turned into a more "faithful" example of ITM, or whether it is just doomed from the start (and why). The changes I suggested don’t sound Irish to me, but more classical (with the implied harmonic minor). Or perhaps what’s missing from it is the ornamentation that would make it sound Irish.

Re: Help needed from experienced session players in or around London

"[*PS - I hope you, the computer and Jim Dorans come up with a nice version of tune #4 *]

Sorry to disappoint you, but we already have :)"

Maybe you’d play it for us.

Posted by .

Re: Help needed from experienced session players in or around London

"perhaps what’s missing from it is the ornamentation that would make it sound Irish"

yes…

..that and altering the entire melody!

Posted .

Re: Help needed from experienced session players in or around London

@irishfiddleCT : What do you suggest? Dead on arrival? Can it be saved? How would you alter it? Which needs more help, the tune or the turn?

Re: Help needed from experienced session players in or around London

I would suggest confining your computerised efforts to the archives of whichever University you are at, I don’t think this helps trad music in any way

I am amazed that these tunes are being given serious consideration by posters on here, whilst there are threads up that criticise the playing of the Kane sisters for instance…

Posted .

Re: Help needed from experienced session players in or around London

[*Maybe you’d play it for us.*]

@Ergo no - *you* need to play it, to decide if it’s any good or not (for you).

Anyway, there’s a link to the tune’s audio at the bottom of each page.

Re: Help needed from experienced session players in or around London

"@Ergo no - *you* need to play it, to decide if it’s any good or not (for you)."

Kind of a generic response, don’t you think? Or haven’t you been reading my posts?

Posted by .

Re: Help needed from experienced session players in or around London

There will be someone to complain about anything.

Bob, I found the work interesting, and I was surprised by how "Irish" the odd tune came out. I didn’t even peruse what would be considered a valid sampling, but if I found 1 tune in the 5 I looked at to be good, it doesn’t seem likely I found the 1 good tune in 3000.

Given the same input, does it return the same set of 3000 tunes every time?

Re: Help needed from experienced session players in or around London

I don’t see the problem with someone playing a tune composed by a computer.
The human brain works like a machine after all - just electrical signals being fired to and fro.
Some may argue that human experience ‘shapes’ a tune but so do many other factors - your own repertoire of tunes, tunes you ‘know’ in your head, music off the telly, your ability to play, etc. Therefore we all use arcane algorithms of some sort to compose tunes, otherwise they wouldn’t sound ‘traditional’. I think the exercise is worthwhile, not so much as a way of improving computer intelligence, but perhaps as a means to understand human intelligence.
I’ve also skimmed some tunes in the list - went almost to the end and worked back. Some were terrible - nice phrases now and again but without a proper resolution, and one or two were actually OK and sounded playable. Pot luck, really. I didn’t find myself saying "these are great tunes" so much as "I could do something with that one", which I think is what really matters. The music comes from the player. A hundred people might play Fishing For Eels, for example, but maybe only one or two play it in a way worth listening to. There are loads of crap tunes written by humans and there will continue to be (as long as there are people like me) in perpetuity. I’d rather play a great tune written by a computer than a crap one I wrote myself.

Re: Help needed from experienced session players in or around London

It seems to me that an interesting experiment would be to take a number of tunes from this experiment and an equal amount of obscure Gan Ainms from the various collections, mix them up with an option to select ‘human’ or ‘computer’, then make that available for trad players to fill out with their best guesses.

Re: Help needed from experienced session players in or around London

"I don’t think Jim Dorans is a walking "dirty bomb" when it comes to session music"

Hey Jim, I hope you plucked that jewel for your press kit! :-P

Re: Help needed from experienced session players in or around London

"does the existence of the 30,000 tunes generated by our system (available in volumes 1-10) affect the value of the real tunes at thesession.org? Or maybe the troublesome feeling is something else. Quality?"

Well, I can only speak for myself, but I already live in a perpetual state of existential angst that I don’t know enough tunes, and the thought that a computer algorithm now exists that could ostensibly crank them out into perpetuity doesn’t exactly take the edge off of The Dread™, if you get me.

Re: Help needed from experienced session players in or around London

@Alex Henry: Thanks for your comments! Our system produces new transcriptions every time we fire it up. We can even give it the first measures of a transcription and have it create the rest. Then we can pick the ones we like, and continue. That is what I do here for well-known Christmas carols: https://highnoongmt.wordpress.com/2015/12/16/tis-the-season-for-some-deep-carols/ I find it a really fun and productive exercise.

@Conán McDonnell: Indeed! We are much more interested in "I could do something with that one" than making any claim that our system is "composing music." More fundamentally, we are interested in questions of what musical knowledge is, and the ways in which our system is modeling it.

@Cheeky Elf: What you describe is known as a "music discrimination toy test." It could be interesting, but our intent in this project is not to fool people. We want to create an intelligent assistant for music composition, education, … who knows! As for your fears, I understand that angst of not knowning enough, and not having enough time on this Earth to know enough. I hadn’t thought about that unintended consequence of our system. :)

Re: Help needed from experienced session players in or around London

Gotcha Bob. I get what you’re doing, and find it very interesting.

A few random observations:

The B part of #1776 sounds familiar but I can’t place it at the moment.

#666 does indeed sound like something Jim Dorans would write.

#1961 sounds more than a little bit like a couple parts of the Bucks of Oramore.

Re: Help needed from experienced session players in or around London

In part it seems that you are hoping the system will get the understand the encoding system.

What would happen if you fed in acoustic information - pitch, duration, volume for starters? There are a lot of other things that a musician generates when playing an intrument that are formalised for MIDI encoding.

Could it then, for example, work out the concept of metre for itself?

I am still not convinced about feeding in messy data that an experimenter has tidied up by inspection; the latter is not very repeatable. There must be several thousand tunes available in ABC that have been well edited. It wasn’t really trained on 23,636 *tunes* was it?

Re: Help needed from experienced session players in or around London

@irishfiddleCT & @Ergo: Thanks for your input. I am sensitive that our work might be seen as a threat to the health of a living tradition. There are certainly ethical issues here! Since a part of my salary is paid by UK tax payers, I am responsible for bridging the gap between my ivory tower and the real world. :) So, please let me offer the following description.

What we are trying to do is to see how the conventions in session music might be modeled statistically through the ABC transcriptions provided by practitioners on thesession.org. How do we know the ways a given model succeeds and fails? One way is to have the model generate new transcriptions and then compare their statistics to those of the real transcriptions. This offers an elementary sanity check, but has limited usefulness because it is disconnected from real world practice. Another way is to have the model generate new transcriptions and ask real practitioners in what ways those transcriptions succeed and fail. This is much more useful for shedding light on what a model has actually learned, and how we can "improve" it. Why even do this in the first place? I make several suggestions above in response to Ergo. Why session music? I really enjoy it, I play this kind of music on my melodeon, I find it a fascinating living music tradition, I believe it deserves to be studied, I believe it can make fruitful contributions to computer science (a good example of art informing science!), and so on.

I hope this helps. In fact, I will condense this discussion into an accessible form for the introduction to each volume of folk-rnn generated transcriptions.

Re: Help needed from experienced session players in or around London

@Cheeky Elf: Thanks for having a look! Regarding 1776, I find that the first measure of the turn appears in the following transcriptions:
Martin Riley’s
The Mountain Road
Not Far From Home
A Laxity Of Morals
A Travers La Vitre
The Five Points
The Green Bridge
The Hesleyside
The House In The Glen
If We Hadn’t Any Women In The World
Jackie Coleman’s

I don’t find in any transcription the first two measures of the turn, even when we remove the triplet.

Interestingly, the first measure of the tune also appears in The Mountain Road. There is some similarity between #1776 and The Mountain Road, but the two are quite distinct looking more long-term.

Thanks for the other observations!

Re: Help needed from experienced session players in or around London

I hear the Mountain Road phrase for sure now, but the rest of the turn is screaming at me, and it’s not matching any of the titles you came up with. Galway Rambler possibly? Are you able to detect the phrase in a different key? This would be so much easier if I had an instrument at hand. The second I played it I’d have the title!

Re: Help needed from experienced session players in or around London

@David50: The original data dump (https://github.com/adactio/TheSession-data) contained 47,924 occurrances of the "T:" field. However, many of these are not complete transcriptions, but comments, or alternative measures, and so on. To clean this up, we removed all transcriptions with less than 7 measures when considering repetitions. We also removed all transcriptions that specify more than one meter or key. This left us with 23,636 *transcriptions*. All the data we used for training comes from here (12.9 MB) https://raw.githubusercontent.com/IraKorshunova/folk-rnn/master/data/sessions_data_clean.txt

As for training a system on audio, that is a much harder problem! In some sense, one wants to create an abstract musically meaningful representation of what can be heard in an audio recording. Automatic music transcription is one route to do this, but is an ongoing research area and unsolved problem because of its high difficulty. We circumvent this by starting from a more ideal situation: what can we do once we have the transcription?

Re: Help needed from experienced session players in or around London

@Cheeky Elf What about these?
Reel De Mattawa
Sheila Coyle’s
Jimmy O’Reilly’s
Knockdhu

Re: Help needed from experienced session players in or around London

@ Bob Sturm. I wasn’t suggesting feeding it the audio. I was suggesting some form of encoding of the audio.

Automatic transcription may be tricky, but I think they have got conversion of a single voice recording into pitch, duration and volume sorted - Audacity will do a graphic representation that is quite good for something like a whistle. Tricky to get that many recordings but some of the ABC to MIDI converters will add emphasis and swing/lilt based on the metre and the dance rhythm. You could generate 23,636 of those. I am thinking of something that is a more direct representation of the sound - a string of parameters analogous to a piano role - that would ‘mechanically’ convert into something a human who knew nothing of music notation would recognise as a tune.

Re: Help needed from experienced session players in or around London

"#666 does indeed sound like something Jim Dorans would write."
Well we all know Jim is one of Satan’s minions.

Re: Help needed from experienced session players in or around London

Tune #667 should be named ‘Neighbour of the Beast’…

Re: Help needed from experienced session players in or around London

[*"#666 does indeed sound like something Jim Dorans would write."
Well we all know Jim is one of Satan’s minions.*]

No, it’s Santa, not Satan, I think.

I always get those two mixed up.

Re: Help needed from experienced session players in or around London

@David50: Thanks for the suggestion. I could see it leading to a multimodal system, taking in acoustic information as well as the transcription!

Re: Help needed from experienced session players in or around London

Jim- it’s the red suit. Mixes me up every time, too.

Re: Help needed from experienced session players in or around London

[*"I don’t think Jim Dorans is a walking "dirty bomb" when it comes to session music"
Hey Jim, I hope you plucked that jewel for your press kit!*]

Well, it would certainly make a change to my usual stream of fan mail :)

Re: Help needed from experienced session players in or around London

Just trying to look out for you, Jim.

Re: Help needed from experienced session players in or around London

If you’re working with seasoned session players around London aren’t most of them able to take any ‘tune’, in any format, give it a listen or a look; decide if it’s more than a random sequence of notes (something worth repeating) or to send another one of thousands back into the bin?

Posted by .

Re: Help needed from experienced session players in or around London

AB: Yes, that would be good, but we are looking for people with whom we can have conversations in person. There has been some interest, but no one has signed up yet. Any ideas on how to incentivise it?

Re: Help needed from experienced session players in or around London

Nomally people would agree a rate and terms if hiring consultants to help develop a system.

Re: Help needed from experienced session players in or around London

@ Beanzy: As academics performing research, not a company developing a product, we have limited resources to hire anyone. We are thinking about other ways, however.

Re: Help needed from experienced session players in or around London

Bob, that is a very good question. I don’t know what is the best incentive. But if I wanted input from experienced session players, myself, I would go to their sessions, listen in and unobtrusively introduce myself. Then, if any of the session players asked for a tune I would beg pardon & comply. I guess what I’m saying is I would just be myself and get to know the musicians I hope to work with.

Does that help?

ps Bob, in your profile there is a link to nearby sessions ~
https://thesession.org/sessions/nearby?latlon=51.52516937,-0.05955100

Posted by .

Re: Help needed from experienced session players in or around London

@AB: Thanks! We have done so at the Duke of Wellington session last month, and plan to go back. I would also like to get a D/G melodeon so I can join in. (Or maybe even go the B/C route … )