Beethoven’s Unfinished Tenth Symphony Gets Completed by Artificial Intelligence: Hear How It Sounds

Few sym­phonies are as well-known as Beethoven’s Ninth, an asser­tion sup­port­ed by the fact that it’s no doubt play­ing in your head even as you read this. Few sym­phonies are less well-known — at least by Beethoven’s stan­dards — than his Tenth, pri­mar­i­ly because he nev­er actu­al­ly got the thing fin­ished. He did make a start on it, how­ev­er, and at his death in 1827 left behind notes and drafts com­posed along­side the Ninth, which had also been com­mis­sioned by the Roy­al Phil­har­mon­ic Soci­ety. Such is Beethoven’s stature that his enthu­si­asts have been spec­u­lat­ing ever since on what his incom­plete sym­pho­ny would sound like if com­plet­ed, employ­ing any tech­niques to do so that their time put at hand.

“In 1988, musi­col­o­gist Bar­ry Coop­er ven­tured to com­plete the first and sec­ond move­ments,” writes Rut­gers Uni­ver­si­ty Art & AI Lab direc­tor Ahmed Elgam­mal at The Con­ver­sa­tion. “He wove togeth­er 250 bars of music from the sketch­es to cre­ate what was, in his view, a pro­duc­tion of the first move­ment that was faith­ful to Beethoven’s vision. Yet the sparse­ness of Beethoven’s sketch­es made it impos­si­ble for sym­pho­ny experts to go beyond that first move­ment.”

When Beethoven’s mile­stone 250th year approached, how­ev­er, the age of arti­fi­cial intel­li­gence was well under­way. To Matthias Röder, the direc­tor of Salzburg’s Kara­jan Insti­tute, unit­ing this tow­er­ing com­pos­er and this promis­ing tech­nol­o­gy had become an irre­sistible propo­si­tion.

Elgam­mal and Röder were just two of the team that came togeth­er to take on the for­mi­da­ble task of engi­neer­ing a form of machine learn­ing capa­ble of help­ing to com­plete Beethoven’s Tenth. The oth­ers includ­ed com­pos­er Wal­ter Wer­zowa (“famous for writ­ing Intel’s sig­na­ture bong jin­gle”), com­pu­ta­tion­al music expert Mark Gotham, and musi­col­o­gist-pianist Robert Levin, who “had pre­vi­ous­ly fin­ished a num­ber of incom­plete 18th-cen­tu­ry works by Mozart and Johann Sebas­t­ian Bach.” Deutsche Telekom pro­vid­ed fund­ing for the project, and also pro­duced the short doc­u­men­tary video on its result above. How­ev­er con­cep­tu­al­ly intrigu­ing, this A.I.-driven musi­cal endeav­or could final­ly be put to the test in only one way: hear­ing it per­formed by a 100-per­cent human orches­tra. As Wer­zowa puts it, look­ing sky­ward, “We hope when he hears it now that he smiles.”

Relat­ed Con­tent:

Stream the Com­plete Works of Bach & Beethoven: 250 Free Hours of Music

Watch Ani­mat­ed Scores of Beethoven’s 16 String Quar­tets: An Ear­ly Cel­e­bra­tion of the 250th Anniver­sary of His Birth

Did Beethoven Use a Bro­ken Metronome When Com­pos­ing His String Quar­tets? Sci­en­tists & Musi­cians Try to Solve the Cen­turies-Old Mys­tery

The Sto­ry of How Beethoven Helped Make It So That CDs Could Play 74 Min­utes of Music

Arti­fi­cial Intel­li­gence Writes a Piece in the Style of Bach: Can You Tell the Dif­fer­ence Between JS Bach and AI Bach?

Based in Seoul, Col­in Mar­shall writes and broad­casts on cities, lan­guage, and cul­ture. His projects include the Sub­stack newslet­ter Books on Cities, the book The State­less City: a Walk through 21st-Cen­tu­ry Los Ange­les and the video series The City in Cin­e­ma. Fol­low him on Twit­ter at @colinmarshall or on Face­book.


by | Permalink | Comments (2) |

Sup­port Open Cul­ture

We’re hop­ing to rely on our loy­al read­ers rather than errat­ic ads. To sup­port Open Cul­ture’s edu­ca­tion­al mis­sion, please con­sid­er mak­ing a dona­tion. We accept Pay­Pal, Ven­mo (@openculture), Patre­on and Cryp­to! Please find all options here. We thank you!


Comments (2)
You can skip to the end and leave a response. Pinging is currently not allowed.
  • Sam says:

    Try­ing to recom­pose beethoven. Wheres,the respect??? Go read a book. He was devine.no sen­si­tiv­i­ty.

  • Isaac Malitz says:

    It’s about a 30% suc­cess: It imi­tates the Beethoven mid­dle-peri­od sound fair­ly well; many of the indi­vid­ual episodes have famil­iar com­po­si­tion­al char­ac­ter. Does not have what I’ll call “Schenker aspect” much either at macro or micro lev­el. Does­n’t have LVB inven­tive­ness, inte­gra­tion. Does­n’t have hooks, explo­sive effects, deep con­trast.. but it is progress for AI. If this were chess, I would say that this AI has emu­lat­ed a play­er in the 800 — 1000 range (“com­pe­tent begin­ner”) but not at all in the 2200+ range (“mas­ter lev­el”). Which is a great accom­plish­ment, a big step for­ward. Look­ing for­ward to future iter­a­tions …

    HOWEVER, two crit­i­cisms

    [a] The AI team does not seem to real­ize that it is only at the 30% lev­el. This is the “Flo­rence Fos­ter Jenk­ins” phe­nom­e­non, where the cre­ator is not aware of the lim­its of their cur­rent scope and capa­bil­i­ty. I know that Levin was involved, and he is a superb musi­cian, but I do not detect his influ­ence in the cur­rent incar­na­tion of the project.

    If the AI team needs a gen­tle cri­tique of their musi­cal work so far, please feel free to con­tact me. I have back­ground and rela­tion­ships that I think could pro­vide use­ful input, in a respect­ful non-dis­rup­tive way. To help you to move for­ward …

    [b] The hype of this project is nau­se­at­ing. It trans­forms a real­ly-good tech­ni­cal effort into a pub­lic fias­co. Again, I think of Flo­rence Fos­ter Jenk­ins, her 1944 Carnegie Hall per­for­mance, where she was high­ly pro­mot­ed AND total­ly beyond her capa­bil­i­ties; she accom­plished one of the great­est fias­cos in music his­to­ry. https://en.wikipedia.org/wiki/Florence_Foster_Jenkins#Vocal_career

    I would encour­age the AI/music cre­ative team to put a leash on their hype-meis­ters. Tell ’em to go away and do some hype on Nora the Cat (“CAT PLAYS PIANO, SIGNS A RECORDING CONTRACT !”)

Leave a Reply

Quantcast