I know the style is based on Bach's four-part chorales, but it sounds more renaissance than baroque to me. I don't know if it's the recorder's sound or the harmony.
Great! I had exactly the same project back in 1985 at Cardiff uni where i was an undergrad. I got the IBM 15xx PC to analyse all the chorales in Riemenschnieder (took a lot of data entry and analysis of vertical and horizontal note relationships and programming (in Pascal i seem to remember). At the end it produced pseudo chorales.
This would be possible. 1-The ai start by generating an 100 random sound of random duration. 2-You give an score to it between 1 and 100. 3-The ai use this score to check how good it did, instead of using other musician music to gauge how good it did. 4-After you rate alot of songs the ai will get better and better, fitting your specific music taste.
I really like this piece. I love how it resolves on a major chord the first time, a minor chord the second time. and another major chord for the final time. However, when it resolves on the minor chord, usually, it would end like ii - V7 - i, but it instead sounded like something else that wouldn't resolve properly. I know this is only the work of the AI, but I think this is important for its development.
much better than i expected. i think it becomes a little less good as it goes on, as it starts to lose direction. but on the level of a bar or two, it is very convincing.
You should upload some kind of "Bach or computer" test on TH-cam, and reveal the results later in another video. People would be more partial without knowing in advance if a chorale is from Bach or DeepBach ;)
It does pretty well. But it's a narrow task. Believe me, when an AI gets as good at creating art as humans (music, stories etc), that would be the threshold for AGI, so there will be a lot more mindblowing world-changing technologies that will make our concern about keeping our jobs seem trivial by comparison.
Bach himself wrote at least seven harmonizations of this choral ('Wer nur den lieben Gott lässt walten"). First of all: he never changes harmonies for the lines 3 and 4, which are the repeated lines 1 and 2. But worse: the harmonies above are miles away from Bach's own compositions and quite amateuristic. For example, there is a horrible quint parallel between soprano and bass in bar no 3.
that's right dude, just explain in the minutest detail what it did wrong so that the developers know how to fix it, you just gave them the solutions, now computers need not ask for help humans will just give them the solution lmao haha praise our new overlords
@@sambulls I've noticed that people get very defensive when it comes to AI. Especially fringe disciplines like a classical music. It's either one of two things (or maybe both): - A severe lack of understanding of how AI or technology progresses. - They're concerned about their career choices being deemed irrelevant by computers. I think he just doesn't understand how research works. He seems to think this is the finished product. This research will probably continue for another decade or so, getting better and better with each iteration. It's inevitable that in time it will progress well beyond human limitations and understanding.
@@Fermion. Except it's trained on human music, so it's never going to surpass it. It's always going to be a mediocre warmed over version of human creativity.
@@philipcai9499 That's not how deep neural networks are trained. For instance, Google's AlphaZero, the current #1 chess entity known to mankind, has never observed even 1 human game of chess. It was simply given the rules. and 4 days later, it was doing things *well* beyond human comprehension by simply playing millions of games with itself. Having it train on human music would introduce a high degree of bias towards that particular composer's style. A true self-learning AI would be given general musical information such as time signature ranges, harmonies and chords, but never a particular composers music.
the tritone B-F in the first notes of the tenor is rarely used in an inner voice like this., and would go down a step after, most likely. also the voice crossing with the alto is not stylistic. this would be more apparent in a vocal performance.
If i have interpretted it correctly , this AI first listens to bach’s music and then recreates it ‘’synthetically’’ to create notes of it trough pattern recognition, now imagine the possibilities, not only can it help music composers to play those songs, but just imagine to be able to convert CD or mp3 music files into synthetic files but identical sounding audio to save on memory space, that would be really cool.
Dude, how was this trained? So much similar motion between the bass and soprano! Machines could, I think, be optimized to do this, but something got missed.
"Wer nur den lieben Gott lässt walten". Already composed by Neumark and used by Bach (and many others). Here we hear a little distorted version. I think the computer is trying to fool us.
Mesmerizing and quite pleasurable. I wonder if it could sustain a longer stretch of music with an overall architecture without becoming incoherent and/or boring.
Try 16th Century England! King Henry VIII wrote music when he wasn't too busy getting divorced. :-P Check out his 'Consort Music' here on TH-cam. Other English Renaissance composers like Thomas Morley, William Lawes, and Thomas Weelkes might be to your liking.
Probably because you associate the sound of recorders with Telemann, who did write a fair bit of music for them. But Telemann wasn't exactly known for his chorale harmonizations!
However interesting, this piece has nothing to do with Bach or the way these settings would be composed by a professional. First and formost, singeing voices are composed in a horizontal way respecting clarity of diction and expression of the text. Bach would choose om the basis of the text structural musical elements and use them in the construction of the setting. There is also an sort of build-up of musical tension from the beginning to the end. Things are interchangeable here. Point is, finding the right chords is the absolute minimum that doesn't make a setting yet. You need more for a cake than flour.
It also has a few mistakes in the sequence of chords, especially suddenly going from major to minor (e.g. the first cadence). But if you don't pay a lot of attention, it does like something Bach composed.
anyway you can try a discrimination test here, especially if you are a Bach expert www.flow-machines.com:3010/ (please disregard the first screenshot that lasts just a few seconds)
Francois Pachet What do you mean? Is this a remake of an actual choral by Bach? And there are indeed cases where he does that, but the only case I can think of being used systematically is when he uses that chord as V/IV and then as I.
Dieser Satz ist sicherlich besser als ein vergleichbarer Satz bei Abschlussprüfungen im Nebenfach Tonsatz an einer Musikhochschule. Jedoch wäre es vermutlich technisch möglich, Oktav- und Quintparallelen zu entdecken und zu vermeiden?
The Bach piece does not work as well as the 'Beatles' piece does ... I assume that's because this is the actual AI footage, without (too much) human interference. It's not Bach. It's not even emotional. It's just a string of music that's not at all impressive because it's just so... generic. Really, the Beatles one was great. But I understand that was actually arranged by a human, this Bach piece is exactly what I'd imagine an AI to be capable of in early 2018. It makes sense, it's musical, it doesn't sound half bad - but it's not even interesting. It's a case study. The very early days of something great to come. Good work, everyone involved; it will get better!
It sounds very nice to me (not an expert), but the first cadence at 0:08 should be major not minor (tenor B natural). There are at least two mistakes in the audio rendition (bass E natural for E flat in bar 2, and soprano mistimed at 0:33). A real chorale would not, I think, leave one part on the same note for so long as happens here in a couple of places.
In fairness, it does plays an E-flat, as written, in bar 2 -- a slightly sharp E-flat, perhaps, but not an E-natural. However, the way it ornaments the soprano line at 0'33" is remarkable, and possibly the one bit of genuine musicality exhibited in the whole exercise!
It's quite good, however I'm guessing this is one of the better samples. How much is chaff? Remember, it's a human that interprets what sounds nice, a human which picked this sample to share with us.
The software is open source and the theory behind is also an open research by many in the field of Neural Networks, If you want to read about it I recommend (It's some kind of easy lecture): karpathy.github.io/2015/05/21/rnn-effectiveness/ To generate "X" composition based on RNN you need to train the RNN with something (Music Notes, Text, etc) and then you can "ask" the RNN to predict / generate new (Music Notes, Text, etc) , there is no "skynet" or theories like that :) The owner of this music is the "trainer" of the RNN, you , me or anyone, maybe Bach haha, ask a lawyer about this part.
The machine (by which I mean its programmers) actually did rather poorly. It ought to have been provided with the basic rules of writing 4-part harmony! (It breaks three or four of them within the first phrase.) Hell, they didn't even tell it how to correctly notate the tenor part! Still, I marvel that the machine couldn't deduce the rules of part-writing and voice-leading from all those flawless examples it studied!
Fermata's...... Bach seldom used them. He would write the rhythm as it was to be played,.... and no this performance does not ignore them,... where are Your sight reading skills?
@@Geopholus Please read exactly. I write about ONE Fermata, not Fermatas. But this is enough. Furtheron this score is not from Bach but from ai. I could read this. Where are You sight reading skills?
@@Geopholus What are you talking about? There are tons of fermata's in Bach's chorale harmonizations. As for the AI ignoring it, it's quite possible that different software was used to play the composition. If that software was not configured correctly it would not pause during a fermata. Somewhat like how a human player could totally ignore a fermata, which would not be the composers fault.
@@0rganopleno< and Martin H Yes, the human performer(s) did not pause on the 2nd to last of 6 fermata's in this 15 measure chorale. It seems from the title here, that the idea of this AI "deep learning", is to generate harmonizations in the style of Bach. This program does a pretty good job of emulating harmonies using traditional voice leading rules of harmonization, but not so much in the style of Bach. Bach seldom used fermata's in his original scores, although just looking through 5 chorals in Bach's hand I do see one that follows a traditional Hymn, that has a regular pattern of many fermatas, while the others covering several hundred measures have 3-5 total and usually only on the final cadence. One notices that the human performer who takes breaths before phrases begin, in this example, chose to ignore the fermata in the 13th measure at his/her/(their) discretion. I don't want AI driving my car, or anybodies car either. But perhaps that is another discussion, especially as I think this is about AI "trying" to compose in the style of Bach, and not the performer(s?) who deleted a fermata. Yes you do see A fermata, and You mean exactly one! Lovely discussion. God forbid, we consider Auto-pilot on aeroplanes or cruise control, or Toyota's accelerate to max software problem several years ago.
@@Geopholus These are human performers? Sounds more like software playback to me. Anyway, it seems to me you were confusing Bach's chorale preludes for organ (or the more elaborate chorale arrangements like in BWV 140 and 147 etc.) and his choral harmonizations. The choral harmonizations have lots of fermata's, mostly because that's where the singers need to breath.
This translation of (Chinese?) to English, shows AI is behind in translation. Is this person saying if Bach had sons they may have written music like this? Then the speculation is wrong, because he did (have sons) and they didn't write music like this. This sounds more like renaissance music,... which sort of makes sense in tha tit is based on a hymn, and it is less complicated rhythmically than Bach. As someone else pointed out there is no building of tension, which is probably the fault of the overall structure to the analysis engine used to construct this AI. If one doesn't know what musical tension is, it would be very difficult to pick up that attribute statistically.
The software is open source and the theory behind is also an "open" research by many in the field of Neural Networks, If you want to read about it I recommend (It's some kind of easy lecture): karpathy.github.io/2015/05/21/rnn-effectiveness/ To generate "X" composition based on RNN you need to train the RNN with something (Music Notes, Text, etc) and then you can "ask" the RNN to predict / generate new (Music Notes, Text, etc) , there is no "skynet" or theories like that :)
Some of the note shifts or sequences are a dead giveaway that it is not Bach. Maybe not a dead giveaway but very unlike Bach. more 19th century, more advanced, more like that of what happens in modern rock and pop as a norm. Bach would more overlay and overlap such things, where the later methods were more along the lines of quick trills up and down the note scale (eg, modern electric guitars in rock and pop or even Chopin, ie high speed virtuosity). Bach would generally achieve this via slight overlap and done via two or more instruments, not one. Via harmonization, if you will. Mozart was more in this direction. I think it has more to do with the instruments and the artists available to play... as well as the underlying societal aspects, of the given times. This thing happens only a few times for a few notes in this composition but it sticks out quite a bit to the ear.
Not bad, but also not gut. Just mediocre. It begins with the first fermata, never in G-Minor, sounds akward. Nevertheless astonishing. A lot of faults you would learn to avoid while studying composing.
this is like if a coder took an intro to music theory class. lol they think they're so deep. you're just beginning to scratch the surface of what actual musicians know.
![Chris Langan - The Interview THEY Didn't Want You To See - CTMU [Full Version; Timestamps]](/img/n.gif)
Machines will be playing this as their anthem while conquering humanity.
Praise
Hilarious, but also creepy as fuck.
I'll oppose them. 🐳 th-cam.com/video/HqUmPKSy0sM/w-d-xo.html
Chances of them actually playing this as their anthem is very low..but never zero lol
Machines will never get as close as Bach's Ricercar a 6 BWV 1079 from Musikalische Opfer
I know the style is based on Bach's four-part chorales, but it sounds more renaissance than baroque to me. I don't know if it's the recorder's sound or the harmony.
@@AMthefox It does have a tint of renaissance, I suppose you could classify it as early baroque.
Yah it's not at all "Bach-like"! I guess that's a good thing.
I think it's the cross-relations in the harmony-the particular way in which is switches between flat and natural E, A, and B isn't always Bachlike
agreed, this is like renaissance italian music where everything just keeps repeating, motivic control aint present
100% agree. More early baroque than Bach.
Great! I had exactly the same project back in 1985 at Cardiff uni where i was an undergrad. I got the IBM 15xx PC to analyse all the chorales in Riemenschnieder (took a lot of data entry and analysis of vertical and horizontal note relationships and programming (in Pascal i seem to remember). At the end it produced pseudo chorales.
I can now imagine an app generating the best tunes you ever heard and never repeating itself.
This would be possible.
1-The ai start by generating an 100 random sound of random duration.
2-You give an score to it between 1 and 100.
3-The ai use this score to check how good it did, instead of using other musician music to gauge how good it did.
4-After you rate alot of songs the ai will get better and better, fitting your specific music taste.
@@exedeath There's no surprise factor in it. Nobody knows how the next hit will sound, and that's why music is so great.
@@slb826 Bach repeats himself a lot. In fact all of his music consists of a few patterns used in creative ways.
@@ymaysernameuay1113 That is called style.
I really like this piece. I love how it resolves on a major chord the first time, a minor chord the second time. and another major chord for the final time. However, when it resolves on the minor chord, usually, it would end like ii - V7 - i, but it instead sounded like something else that wouldn't resolve properly. I know this is only the work of the AI, but I think this is important for its development.
much better than i expected. i think it becomes a little less good as it goes on, as it starts to lose direction. but on the level of a bar or two, it is very convincing.
You should upload some kind of "Bach or computer" test on TH-cam, and reveal the results later in another video. People would be more partial without knowing in advance if a chorale is from Bach or DeepBach ;)
It's very easy to tell 😂😂
There goes my job.
They took er jobs.
haha "THEY BROKE HIS JAW !"
you 'disintegration' will always be my favorite album of all time.
your job is to write shitty, short chorale harmonizations? you should be glad to avoid doing that anymore
It does pretty well. But it's a narrow task. Believe me, when an AI gets as good at creating art as humans (music, stories etc), that would be the threshold for AGI, so there will be a lot more mindblowing world-changing technologies that will make our concern about keeping our jobs seem trivial by comparison.
Bach himself wrote at least seven harmonizations of this choral ('Wer nur den lieben Gott lässt walten"). First of all: he never changes harmonies for the lines 3 and 4, which are the repeated lines 1 and 2. But worse: the harmonies above are miles away from Bach's own compositions and quite amateuristic. For example, there is a horrible quint parallel between soprano and bass in bar no 3.
that's right dude, just explain in the minutest detail what it did wrong so that the developers know how to fix it, you just gave them the solutions, now computers need not ask for help humans will just give them the solution lmao haha praise our new overlords
@@sambulls I've noticed that people get very defensive when it comes to AI. Especially fringe disciplines like a classical music. It's either one of two things (or maybe both):
- A severe lack of understanding of how AI or technology progresses.
- They're concerned about their career choices being deemed irrelevant by computers.
I think he just doesn't understand how research works. He seems to think this is the finished product. This research will probably continue for another decade or so, getting better and better with each iteration. It's inevitable that in time it will progress well beyond human limitations and understanding.
I also highly doubt that the part on the bass at the end would be done by deep learning.
@@Fermion. Except it's trained on human music, so it's never going to surpass it. It's always going to be a mediocre warmed over version of human creativity.
@@philipcai9499 That's not how deep neural networks are trained. For instance, Google's AlphaZero, the current #1 chess entity known to mankind, has never observed even 1 human game of chess.
It was simply given the rules. and 4 days later, it was doing things *well* beyond human comprehension by simply playing millions of games with itself.
Having it train on human music would introduce a high degree of bias towards that particular composer's style. A true self-learning AI would be given general musical information such as time signature ranges, harmonies and chords, but never a particular composers music.
its like some notes that sound good but go nowhere
maybe this is DeeperBach th-cam.com/video/HqUmPKSy0sM/w-d-xo.html :)
I totally agree. When I listen to it, it feels like reading a book with random content...
That's a good observation. A computer can replicate emotional music, but how can it truly create and replicate what it cannot feel or experience?
I dont think you understand BACH as well as you think. This was also 2 years ago, a lot has changed since.
@@14000091 Bach the composer, or the program you mention? I have no knowledge of the computer's abilities other than what I've seen.
wow, gorgeous !!, I can't believe this piece is from an AI.
time to put my 980ti to good use
I was thinking of putting my C64 to good use.
interesting, however this sounds rather as a renaissance composition (motet), the very first bars absolutely don't sound as Bach ( harmonies )
the tritone B-F in the first notes of the tenor is rarely used in an inner voice like this., and would go down a step after, most likely. also the voice crossing with the alto is not stylistic. this would be more apparent in a vocal performance.
Still long way to go. (Wait a minute. Is this video 8 years ago?)
Holy shit, I didn't realize until I read your comment
Pretty good. I could see that being a chorale by Bach.
If i have interpretted it correctly , this AI first listens to bach’s music and then recreates it ‘’synthetically’’ to create notes of it trough pattern recognition, now imagine the possibilities, not only can it help music composers to play those songs, but just imagine to be able to convert CD or mp3 music files into synthetic files but identical sounding audio to save on memory space, that would be really cool.
yea it seems valuable as a tool to propose options for the next verses to music producers
For years I wondered: when will it come out for all of us to actually try/use it? I suppose that will never happen, right?
This is pretty nice!
Cool, wish it can bring us more beautiful songs!
Prepare to meet your computer overlords
Dude, how was this trained? So much similar motion between the bass and soprano! Machines could, I think, be optimized to do this, but something got missed.
"Wer nur den lieben Gott lässt walten". Already composed by Neumark and used by Bach (and many others). Here we hear a little distorted version.
I think the computer is trying to fool us.
Fully agree. In fact, a traditional medieval French tune, "une jeune fillette", seems to be the common stem.
this should be used for a scene in a movie just basking in --the glory of a-- destructed society.
Mesmerizing and quite pleasurable. I wonder if it could sustain a longer stretch of music with an overall architecture without becoming incoherent and/or boring.
I want more, when will we have more of this nice music?
Try 16th Century England! King Henry VIII wrote music when he wasn't too busy getting divorced. :-P Check out his 'Consort Music' here on TH-cam. Other English Renaissance composers like Thomas Morley, William Lawes, and Thomas Weelkes might be to your liking.
That parallel fifth in the second and third beats of the first bar though.
And later on in the chorale too (btw, love your avatar)
These are not actually parallel fifths, as the first interval is a diminished fifth and the second is a perfect fifth -- allowable, if a little crude.
wow i loved it!
attention aux Octaves parallèles..
Deep respect!
Strikes me as more like Telemann.
But - the fact that it strikes me as "like" anyone is quite impressive.
Probably because you associate the sound of recorders with Telemann, who did write a fair bit of music for them. But Telemann wasn't exactly known for his chorale harmonizations!
Why is the rhythm different from the score at 00:15 / 00:16? In any case, very interesting! (though it doesn't quite sound like Bach yet IMHO)
One can identify the part in the beginning up to 00:43 as part of an actual chorale written by Bach himself, so what exactly did the AI do?
the important point is whether the AIs listening to it like it or not
That;s still not Bach, but it's damn cool.
It is very similar to BWV 84 choral but it's still have some differences.
i wonder how do you extend the bars in nonoto, i am using that interface. but the preset limit of 10 bars cant even generate a convincing motif
However interesting, this piece has nothing to do with Bach or the way these settings would be composed by a professional. First and formost, singeing voices are composed in a horizontal way respecting clarity of diction and expression of the text. Bach would choose om the basis of the text structural musical elements and use them in the construction of the setting. There is also an sort of build-up of musical tension from the beginning to the end. Things are interchangeable here.
Point is, finding the right chords is the absolute minimum that doesn't make a setting yet. You need more for a cake than flour.
It also has a few mistakes in the sequence of chords, especially suddenly going from major to minor (e.g. the first cadence). But if you don't pay a lot of attention, it does like something Bach composed.
anyway you can try a discrimination test here, especially if you are a Bach expert www.flow-machines.com:3010/ (please disregard the first screenshot that lasts just a few seconds)
going from major to minor is also done by Bach in the original chorale
Francois Pachet What do you mean? Is this a remake of an actual choral by Bach? And there are indeed cases where he does that, but the only case I can think of being used systematically is when he uses that chord as V/IV and then as I.
It is a reharmonisation of a chorale "Wer nur den lieben Gott lasst walten" for which we just kept the soprano.
Dieser Satz ist sicherlich besser als ein vergleichbarer Satz bei Abschlussprüfungen im Nebenfach Tonsatz an einer Musikhochschule. Jedoch wäre es vermutlich technisch möglich, Oktav- und Quintparallelen zu entdecken und zu vermeiden?
Can this generative model be conditioned on input observations, such as landscape pictures, in order to simulate a kind of inspirational environment?
Some ornaments too many- actually sounds like a recorder quartett, like "wer nur den Lieben Gott lässt walten" or am I wrong?
Not Bach style. Sounds like a bootleg version of Frescobaldi
It would have been better to use a picardy 3rd chord at the cadence.
in a work with no effort
there will bo no soul
The Bach piece does not work as well as the 'Beatles' piece does ... I assume that's because this is the actual AI footage, without (too much) human interference. It's not Bach. It's not even emotional. It's just a string of music that's not at all impressive because it's just so... generic. Really, the Beatles one was great. But I understand that was actually arranged by a human, this Bach piece is exactly what I'd imagine an AI to be capable of in early 2018. It makes sense, it's musical, it doesn't sound half bad - but it's not even interesting. It's a case study. The very early days of something great to come. Good work, everyone involved; it will get better!
Sounds like my first few chorales
I bet you would buy it if someone said that was original Bach
It sounds very nice to me (not an expert), but the first cadence at 0:08 should be major not minor (tenor B natural). There are at least two mistakes in the audio rendition (bass E natural for E flat in bar 2, and soprano mistimed at 0:33). A real chorale would not, I think, leave one part on the same note for so long as happens here in a couple of places.
0:08 was the first thing that struck my ears as well...It's just OKAY otherwise, lol
In fairness, it does plays an E-flat, as written, in bar 2 -- a slightly sharp E-flat, perhaps, but not an E-natural. However, the way it ornaments the soprano line at 0'33" is remarkable, and possibly the one bit of genuine musicality exhibited in the whole exercise!
Test is unavailable now :(
Really wanted to try it out.
It's quite good, however I'm guessing this is one of the better samples. How much is chaff?
Remember, it's a human that interprets what sounds nice, a human which picked this sample to share with us.
Interesting approach. Was there a particular reason to avoid the use of rests?
@ignacio-araya that's how Bach chorales are usually
I think bachs style has more interplay between the various melodies
I've had it.
go here because of the google doodle bach machine learning
Deepbach is better than the Google doodle
Sorry but this is really just the Chorale from BWV 93 in a different key with some slight rhythm adjustments and decorative notes. Overfitting?
Anyone here from TwoSetViolin? 🤗
made me think of monteverdi actually
Would love to see a generated fugue
IMPRESSIVE!!
Sounds like music from Runescape, neat
Yes, exactly. Twoset brought me here too
alguem da PUC-RJ do curso de AI-I ? lá esse audio estava mudo rsrsrs
For the sake of fairness it should be at least mentioned that it's all just a harmonization of an already existing choral.
Link 404'd?
sorry we just changed the URL. This is the new one:
www.flow-machines.com/deepbach-polyphonic-music-generation-bach-chorales/
What program for playback?
So, sky net will want us to like good music? And those who will not - should be terminated.
This feels more renaissance than baroque - Bach would have used much more counterpoint
Sounds a bit like medieval polyphony
Exactly !
machine needs to have emotions to emulate directions without that it's all random and sparkly stones
Parallel octaves between bass and soprano? AI has still a lot to learn 😀
... but it is still Bach in a way, because it have learned from him, right? :)
So, who is the legal owner of this music if this software can be purchased?
The software is open source and the theory behind is also an open research by many in the field of Neural Networks,
If you want to read about it I recommend (It's some kind of easy lecture):
karpathy.github.io/2015/05/21/rnn-effectiveness/
To generate "X" composition based on RNN you need to train the RNN with something (Music Notes, Text, etc) and then you can "ask" the RNN to predict / generate new (Music Notes, Text, etc) , there is no "skynet" or theories like that :)
The owner of this music is the "trainer" of the RNN, you , me or anyone, maybe Bach haha, ask a lawyer about this part.
very creepy thing.... its realy amazing but still so dangerous work of very clever people
The machine (by which I mean its programmers) actually did rather poorly. It ought to have been provided with the basic rules of writing 4-part harmony! (It breaks three or four of them within the first phrase.) Hell, they didn't even tell it how to correctly notate the tenor part! Still, I marvel that the machine couldn't deduce the rules of part-writing and voice-leading from all those flawless examples it studied!
I see a fermata - he ignores it. I don't want to sit in a car driven by ai.
Fermata's...... Bach seldom used them. He would write the rhythm as it was to be played,.... and no this performance does not ignore them,... where are Your sight reading skills?
@@Geopholus Please read exactly. I write about ONE Fermata, not Fermatas. But this is enough.
Furtheron this score is not from Bach but from ai. I could read this. Where are You sight reading skills?
@@Geopholus What are you talking about? There are tons of fermata's in Bach's chorale harmonizations. As for the AI ignoring it, it's quite possible that different software was used to play the composition. If that software was not configured correctly it would not pause during a fermata. Somewhat like how a human player could totally ignore a fermata, which would not be the composers fault.
@@0rganopleno< and Martin H Yes, the human performer(s) did not pause on the 2nd to last of 6 fermata's in this 15 measure chorale. It seems from the title here, that the idea of this AI "deep learning", is to generate harmonizations in the style of Bach. This program does a pretty good job of emulating harmonies using traditional voice leading rules of harmonization, but not so much in the style of Bach. Bach seldom used fermata's in his original scores, although just looking through 5 chorals in Bach's hand I do see one that follows a traditional Hymn, that has a regular pattern of many fermatas, while the others covering several hundred measures have 3-5 total and usually only on the final cadence. One notices that the human performer who takes breaths before phrases begin, in this example, chose to ignore the fermata in the 13th measure at his/her/(their) discretion. I don't want AI driving my car, or anybodies car either. But perhaps that is another discussion, especially as I think this is about AI "trying" to compose in the style of Bach, and not the performer(s?) who deleted a fermata. Yes you do see A fermata, and You mean exactly one! Lovely discussion. God forbid, we consider Auto-pilot on aeroplanes or cruise control, or Toyota's accelerate to max software problem several years ago.
@@Geopholus These are human performers? Sounds more like software playback to me. Anyway, it seems to me you were confusing Bach's chorale preludes for organ (or the more elaborate chorale arrangements like in BWV 140 and 147 etc.) and his choral harmonizations. The choral harmonizations have lots of fermata's, mostly because that's where the singers need to breath.
model.add(Dropout(0.3))
Oh my God.
So A.I. composed this music?!
But it's still just a human invention.
if Bach's bad son of the can compose, this will be doing.
This translation of (Chinese?) to English, shows AI is behind in translation. Is this person saying if Bach had sons they may have written music like this? Then the speculation is wrong, because he did (have sons) and they didn't write music like this. This sounds more like renaissance music,... which sort of makes sense in tha tit is based on a hymn, and it is less complicated rhythmically than Bach. As someone else pointed out there is no building of tension, which is probably the fault of the overall structure to the analysis engine used to construct this AI. If one doesn't know what musical tension is, it would be very difficult to pick up that attribute statistically.
Se parece algo.. pero hay cosas que bach no haría como la falsa relación del 24''
This is beautiful...
Machines have better feelings than humans ?
They can be more humans than many humans
:D
Wrong! It's in the style of the renaissance era.
Agreed!
i was brought forth by twoset
Sounds like Carcassonne music.
Sounds like bad Telemann. At best.
OMG I am so scared of this
*Well, that wasn't bad, but it's pure Renaissance...not Bach or even Baroque music as a whole.*
is it open source?
The software is open source and the theory behind is also an "open" research by many in the field of Neural Networks,
If you want to read about it I recommend (It's some kind of easy lecture):
karpathy.github.io/2015/05/21/rnn-effectiveness/
To generate "X" composition based on RNN you need to train the RNN with something (Music Notes, Text, etc) and then you can "ask" the RNN to predict / generate new (Music Notes, Text, etc) , there is no "skynet" or theories like that :)
0:27 uh...
没用巴赫内味儿，更像 Telemann 的
Bach > IA (Many issues in the score :/ )
pause at 0:14. look at alto and bass, between third and fourth meter. parallel octaves! Bach would never do that! haha
Very nice. I like this choral. Wer nur den lieben gott lässt walten :))
魂の抜けたバッハ感、確かにバッハバッハしてるんだけどもこれはうーん
Some of the note shifts or sequences are a dead giveaway that it is not Bach. Maybe not a dead giveaway but very unlike Bach. more 19th century, more advanced, more like that of what happens in modern rock and pop as a norm. Bach would more overlay and overlap such things, where the later methods were more along the lines of quick trills up and down the note scale (eg, modern electric guitars in rock and pop or even Chopin, ie high speed virtuosity). Bach would generally achieve this via slight overlap and done via two or more instruments, not one. Via harmonization, if you will. Mozart was more in this direction. I think it has more to do with the instruments and the artists available to play... as well as the underlying societal aspects, of the given times. This thing happens only a few times for a few notes in this composition but it sticks out quite a bit to the ear.
Not bad, but also not gut. Just mediocre. It begins with the first fermata, never in G-Minor, sounds akward. Nevertheless astonishing. A lot of faults you would learn to avoid while studying composing.
this is like if a coder took an intro to music theory class. lol they think they're so deep. you're just beginning to scratch the surface of what actual musicians know.
Prob toomore time for human to 'tune' this to get.it
The difference with real Bach is that here nothing happens througout the whole piece. A huge difference!
It's more modal than Bach. Sounds more like Tallis, or Byrd.
12 Bots disliked this