Holly+ Combines Human-based AI Voice, Harmonizing Algorithm to Recreate Hit Pop Song
A story on engadget.com introduces the world to Holly, a singing algorithm derived from the voice of experimental singer Holly Herndon. Holly Herndon released an album last year called Proto, which co-creator Mat Dryhurst produced. Together they developed an AI-driven voice algorithm called Spawn, “a singing neural network” created through machine learning.
Recently, Holly released a single track remaking the Dolly Parton hit song “Jolene.” The vocal harmony on the track is done by Holly + an algorithm. Holly+, created in partnership with Never Before Heard Sounds, debuted to the public last year, allowing anyone to use a model of Holly’s voice. Now she’s released a new single, where the only vocals are her digital twin.
A video of the song below demonstrates how the algorithm and the AI-generated images work together. The animated hair braid is reminiscent of Pippi Longstocking’s squirrelly locks, though Holly’s seems to have a mind of its own.
Holly does a digital version of Dolly Parton’s big hit ‘Jolene.’
Writer T. O’Brien discussed the quality of the song how and it came about, explaining that the soundtrack is a faithful reproduction of the original, if slower, and “Ryan Norris, who handles the instrumentation, doesn’t take any extreme liberties with the arrangement or sound palette.”
“It simply swaps frantic desperation with plaintive resignation,” T. O’Brien writes.
He points out that some critics said they found the track annoying, claiming that Holly has been given a body for the video that only an algorithm could love. Moving stiffly, Holly is constantly shifting around, and her dancing braid is distracting. Most major artists’ experimentations with AI have generated soundscapes and synth melodies. The article writer said he thinks this is “the first time a machine-learning model has taken the mic solo in a pop song.”
O’Brien also mentions “stilted phrasing” and “digital artifacts” that make the song sound robotic, however.
There is no denying how truly magical the progress these algorithms are in generating digital images. Now the question is can they do the same with audio generators? How machine learning taught an algorithm how to harmonize is a stunning feat.
Perhaps go back and listen to it again without watching the video. It gives you a better idea of where there is still work that needs to be done if Holly hopes to crack the Top 10 of the country music charts.
Perhaps the biggest question was how would AI create feelings of need or jealousy. And how would AI choose to deal with heartbreak? We know from experience a variety of emotions can overwhelm a human heart trying to deal with love. It should be fascinating to see how machine learning tries to solve that “crazy little thing called love” problem.
read more at engadget.com