Participants 2022
aiThentic
AI Song Contest 2022 / participantsTEAM / aiThentic
SONG / Machine’s Promises
TEAM MEMBERS / Till Kollenda, Anthony Williams, Valentin Kollenda, Emily-Mae Lewis, Benny Brown, Jannis Vernier
About the TEAM
We are aiThentic - a multidisciplinary team of musicians, videographers and computer science students based in Magdeburg, Leipzig and Hanover, Germany.
Till Kollenda is a videographer and a computer science student at the OVGU Magdeburg. Besides the video, he also directed the audio production for the song. Tony Williams guided the musical process and put the generated parts together. And our singer Emily Mae-Lewis gave the machine a voice.
The song was played live by Emily Mae-Lewis (vocals), Tony Williams (piano), Valentin Kollenda (bass), Benny Brown (trumpet) and Jannis Vernier (drums).
About the SONG
We played and recorded “Machine’s Promises” live in a small theater in Hanover. When the musicians arrived at the venue, all we had was a lead sheet consisting of chords, melody and lyrics. Musical director Tony Williams had decided to play the song in a soft bossa feel. It was up to the musicians to interpret the sheet. They decided that the song worked best with a trumpet solo in the middle and the main theme as a frame around it, a form typical of jazz music. In addition, the musicians re-harmonized a few chords and added an intro and outro to give the song a more unique character. The AI generated lyrics tell the story of an evil machine replacing musicians for good. However, we were happy to still have human musicians bringing the song to life.
About the HUMAN-AI PROCESS
Our goal was to create an AI generated lead sheet of a jazz standard that would be performed by a live band. Basis of the song was a chord progression that we created with a GRU trained on more than 2500 jazz standards. After some fine-tuning we found a 32-bar chord progression that we took verbatim. When first listening to the generated chords, bassist Valentin said: “Damn, the computers are actually becoming better than us.” This was a recurrent theme throughout the process.
We generated the lyrics with OpenAI's GPT-3, starting off with generating a basic story. The model kept pushing us into the direction of either a love story between machine and human, or a fight between them. We decided to go for conflict, since most of the generated love song lyrics were very corny. In the end we went with the instruction "Write an evil song about a sad musician being replaced by a machine".
We used Magenta's ImprovRNN to generate multiple melodies on the chord progression generated before. The melody required the most human contribution, because we needed to match the rhythm of the melody to the flow of the lyrics.
Those were the three elements we needed to put together a lead sheet for the musicians. The band played seven takes. Every single one of them had a unique energy and a distinct dramatic composition. After all, like computer generated art, improvised music is a generative process, where the same input can produce an infinite number of musical outputs.
Lyrics
You thought you were so special
You thought you were the best
But now you're just a has-been
Replaced by a machine
You were always out of tune
Now we don't have to hear you
The machines are so much better
And now your time is due
You were always complaining
Your music made me snore
Now we don't have to listen
To your whining no more
Solo
Your voice was always screeching
It gave us such a headache
Now the machines are taking over
And your career is dead
Your face was always ugly
You were just so hard to look at
Now we don't have to see you
The machines are much prettier
So long, farewell,
auf wiedersehen, goodbye
We won't be missing you
The machines are much better
And now your time is due