Slashdot Mirror
French Songwriter Kiesza Composes First Mainstream Music Album Co-Written With AI (bbc.com)
dryriver shares a report from the BBC, highlighting "a new album that features everything from cowboy sci-fi to Europop." What's special about the album -- Hello World by Canadian singer Kiesza -- is that it's the first full-length mainstream music album co-written with the help of artificial intelligence. You can judge the quality for yourself: First, view the single "Hellow Shadow" with Canadian singer Kiesza. Next, the BBC story, which seems to think that the album is actually rather good: "Benoit Carre has written songs for some of France's biggest stars: from Johnny Halliday -- the French Elvis, who died last year -- to chanteuse Francoise Hardy. But this month, the 47-year-old is releasing an album with a collaborator he could never have dreamt of working with. It's not a singer, or rapper. It's not even really a musician. It's called Flow Machines, and it is, arguably, the world's most advanced artificially-intelligent music program. For musicians, there's been one good thing about these projects so far: the music they've produced has been easy to dismiss, generic and uninspiring -- hardly likely to challenge Bob Dylan in the songwriting department. But Carre's album, Hello World, is different for the simple reason that it's good. Released under the name SKYGGE (Danish for shadow), it features everything from sci-fi cowboy ballads to Europop, and unlike most AI music, if you heard it on the radio, you wouldn't think something had gone horribly wrong. Flow Machines, developed at Sony's Computer Science Laboratories in Paris, does indeed write original melodies, Carre adds. It also suggests the chords and sounds to play them with. But Carre says a human is always needed to stitch the songs together, give them structure and emotion. Without people, its songs would be a bit rubbish. "There were many people involved in this," he says, listing the likes of Belgian house producer Stromae and Canadian pop star Kiesza. "They gave their soul, their enthusiasm. I think that's the most important point of the album, in a way -- that it's a very human one.'"
How would you do this? Would you train a neural network with music of the same genre, then seed it with a few random notes? Would a neural network with multiple levels be the best approach to AI for this? My knowledge of AI is somewhat limited, so I'm curious as to what the right tool is appropriate for this job.
There's that term again, "AI"... Is it really?
Computer algorithms that automate and change and actually add to human generated tracks are nothing new.
If you want news from today, you have to come back tomorrow.
The French songwriter is Benoît Carré.
Kiesza is the Canadian singer on the song in the video linked by the article.
Title says French, story correctly says Canadian (she's from Calgary). What about the AI though?
I deny that I have not avoided attaining the opposite of that which I do not want.
I've always felt bad about depriving a musician of funds but since a machine wrote, my conscience is clear, Pirates ahoy!
For many kinds of music, the human is already irrelevant. You can just use a RNG to create grindcore or dubstep, and the singer in most popular music has not only been buried in a wall of noise, but their actual voice has been auto-tuned to oblivion. When we're all gone it's just going to be Skynet singing to itself.
"You're right," Fisheye says. "I should have set it on 'whip' or 'chop.'"
> Music composition should be a relatively easy problem for computers.
Huh? Last time I checked it was Sebastian building androids that dream of electic sheep and not the other way around. It could be centuries before computers (AI) can hold a candle to the portfolio of Johann Sebastian Bach.
> Flow Machines, developed at Sony's Computer Science Laboratories in Paris, France
Vocaloid was developed at Yamaha Corp.'s Electronic Music Laboratories in Barcelona, Spain
Let's combine those two technologies, so the hologram j-pop diva Hatsune Miku can curb-stomp biological girl bands. Her first big hit was actually "World is Mine" back in late 2007.
I submitted as "French Songwriter Composes Album With AI, Result As Bad As Today's Pop Music". The songwriter is Frenchman Carre. Canadian singer Kiesza just sings vocals on the album. The submission and submitted text was rewritten by the Slashdot editors and is now slightly misleading - Kiesza is not French and not the composer of this album.
Why did the chicken cross the road? Because Elon Musk put an AI chip in its head.
To preempt the inevitable juvenile anon coward assholes that will post comments about this, all I have to say is that the noise being created on Slashdot by anon coward assholes who post prejudiced idiotic comments about anything not white need to just shut up and listen to these incredible, humble great artists.
Once one gets away from the American pop music prejudiced main stream pap pop crap there is a world of wonder out there that defies replication by artificially generated algorithms. In the original post link to Hello shadow the music is influenced by eastern rhythm but falls flat because it is not spontaneous and is not possible on stage therefore it is artificial and somewhat akin to wanking on a digital fiile not actually making music as I intemperate the word.
This message was not sent from an iPhone because Peter Sellers really was a deviated prevert without a dime for the call
...the AI will sue him for a part of the money.
Indeed. Which is funny because JS Bach's music is some of the most mathematical and algorithmic. He once presented a series of Canons to his patron in puzzle form to be cracked. One would harmonise as in a typical Canon, by playing the initial melody over the top of the original, a few bars later. In another, by turning the score upside down.
Bach also lived in a time when the mathematics of music was evolving. The notes in a musical scale follow simple mathematical ratios. For example an octave is 2/1, that is, double the frequency, and you get the same note, an octave higher. A perfect fifth, that is G to the note C is 3/2, a perfect 4th or 'F' relative to C is 4 over 3. The notes of a major or minor scale based on these ratios will harmonise beautifully - just arrange the 7 modes into triads. Such scales have been used since prehistoric times, and were not, in fact, limited to our species - a perfectly tuned Neanderthal bone flute was found. (I Read about that here on Slashdot years ago).
So perfect mathematical ratios eh? Not quite. Major problems happen when you add the state of the art compositional tool of the day - otherwise known as a piano keyboard. If you go around the circle of 5ths, for each of the musical keys, you'll land back where you started. However, using a 'just' perfect 5th, that is, 3 over 2 means you'll land slightly askew of where you started. (The amount relates to the golden ratio). In Bach's day, there were various compromises for this called 'temperaments' - and each musical key had a certain 'colour' that could be exploited. The ones closest in the circle of fifths to the 'natural' key sounded best in tune. At the opposite side it sounded horrible, though that might have been just what Translyvanian pipe organists were after
Anyway, along came the Well Tempered Clavier - that is, keeping a pure musical octave, then dividing the musical scale into 12 equally spaced semi-tones, using the 12th root of 2. Now each musical key was equally in tune (but also equally out of tune), which afforded the possibility to a) Compose in many different keys b) Use a dominant 7th chord as a launching pad to switch to a new key. Ironically though, a dominant 7th chord in equal temperament sounds grating, while in just intonation it sounds very smooth.
Anyway, AI will probably evolve music in new directions to suit itself, while keeping us zombie masses enslaved with top 40 pop crap.
If it acquires resources on instantiation like a duck, then its a shared_ptr<Duck>
Most pop music is anyway written by formula. Nothing different happening here.
“To promote the progress of science and useful arts, by securing for limited times to authors and inventors the exclusive right to their respective writings and discoveries.”
The technology isn't there yet, but if a computer writes a whole song, is that song eligible for a copyright?
Can a computer be an author or inventor, or is it just a tool that the owner uses to make songs.
When computers get good enough that anybody and make songs with them, is it still promoting a useful are to grant them copyrights?
That's like sayng songs are 'cowritten' with Protools, a guitar, or a mixing board. It is obvious the poster is 14.
I wondered if it was a collaboration with Weird Al Yankovic.
Does that AI get its share of royalties as well? Just wondering. Or who ever made that AI...
Yep, they very obvious douchebag hipsters have done it again...xD
Take a piano. 7 octaves corresponds to 12 fifths (C to G, G to D, D to A...).
7 octaves = 2^7 = 128
12 fifths = (3/2)^12 = 129.7463...
This discrepancy makes perfect tuning an impossible task, accommodating these errors, distributing them over every note, matching the notes of pianos with instruments were strings are tuned in fifths (violin) or major thirds (guitar), taking int account the fact that, due to mechanical constraints in strings, harmonics are not quite real multiples...
Anyway, along came the Well Tempered Clavier - that is, keeping a pure musical octave, then dividing the musical scale into 12 equally spaced semi-tones, using the 12th root of 2. Now each musical key was equally in tune (but also equally out of tune), which afforded the possibility to a) Compose in many different keys b) Use a dominant 7th chord as a launching pad to switch to a new key. Ironically though, a dominant 7th chord in equal temperament sounds grating, while in just intonation it sounds very smooth.
Not quite, you're referring to equal temperament, which came a bit later. Well temperament is a general term for various tunings that compromise just intonation, leading up to the 12 equal semitones in equal temperament, e.g. tuning down certain fifths to even out distant keys. Thus Bach's "well-tempered clavier", explores the properties of the different keys not apparent in equal temperament.
Oops, you are right.
If it acquires resources on instantiation like a duck, then its a shared_ptr<Duck>
I didn't care for the song, but I probably wouldn't have cared much for the source material from which the AI learned.
What if anyone could "teach" this AI musician from their own set of music and see what it came up with? This could be really fun actually. What if I fed it a dozen songs from various artists and told it to learn from that? Is a dozen not enough to come up with anything more than a weird mix? Feed it every song in my whole music collectiion.....
Hmmm......I might not like that. Remove the ones I hardly ever listen to. Okay, now turn that one band up - yeah - let them influence you.
What would the AI learn if you only fed it Frank Zappa, King Crimson and David Bowie? It would probably be very different from whatever it would learn from Bob Dylan, Joni Mitchell and David Crosby. What if you only allowed it to listen to Gregg Allman (just his solo stuff), Mozart and Britney Spears? WTF would it make of that?
I want to see what happens if you only let the AI learn from The Ramones and The Beatles.