Summary of the various ideas that came together in the music that we made for the song contest.

Music setup at OVGU AILab, Magdeburg

AILab at OVGU, Magdeburg

This is where we spent most of our time making music with our AI tools! Most of our tools are within the computer, but we created a basic setup to record and jam with equipment we found at the lab.

Our song is inspired from a story generated by a GPT-2 model. This GPT-2 model was trained on lyrics from famous pop songs. We generated lots of pieces, but this one in particular caught our interest. So we decided to adopt this story and turn it into a song!

Main story (GPT2 trained on this dataset):

The legends and the myths

There’s a girl who can fly

And the wise one will guide her home

The dragon she ran from was defeated

The dragon she rolled from her jaws

The dragon she fought from was slain

Just a month’s journey from home

And now she’s out of sight

Oh love will guide it

So lost and out of mind

Just a month’s journey

And now you’re out of mind

Now you’re out of mind

I wonder where you’re going

You’re all I need

And if I get far I will go

To make up my mind

Don’t let me go and let love guide it all all

Come on and give it all you’ve got

Parallely, we were also experimenting with various AI tools capable of generating MIDI and audio. To create the various musical elements, we used tools open sourced by the Google Magenta Project. All drums, percussion, basslines and melodies are created by passing outputs between the various apps within Magenta studio. We use the Ableton Max for Live Device for Magenta studio.

Since the story is about a journey, we also thought of placing this story near an ocean. To add the sound of the ocean using AI tools, we used NSynth. We take two sounds - intro of the song and sound of the ocean. We then encode these sounds using the NSynth Encoder, cross-fade them in encoding space and resynthesize to create a cross-fading of the two original signals with AI artifacts. We found them pretty cool, so we added them at the start and end of the song.

In order to preserve a tropical vibe in the song, we try to use shaker hi-hats, bongos, congos and chilled out drum set, all played with outputs from Magenta studio. The chord progression was decided by us and was recorded through a MIDI controller. We spent a lot of time jamming to various basslines and melodies with our guitar! For the melodies, we use percussive plucks layered with airy noise instruments and flute-like instruments.

None of us are confident singers on record, so we found it hard to record vocals. Instead of singing much, we tried to give voice to our favorite parts of the AI generated text. We record four lines from the original story. Then we pass them through DDSP models trained on the NUS-48E dataset. Each of these DDSP models is trained on a specific singer in the dataset. We then resynthesize new versions of the original recording from these models and use them to layer the original vocals with AI artifacts. The DDSP models work with a sample rate of 16000 Hz, which makes their resynthesis sound fuzzier compared to the original recording recorded at 44100 Hz.

Then we add in some risers, a 4/4 kick drum and necessary sound effects while also tweaking small parts of the sound to ensure that the track has good listening quality.

The final lyrics for the song ended up being -

Just a month’s journey from home

And now she’s out of sight

Oh love will guide it

So lost and out of mind

And thats why, we call this musical piece Love Will Guide It.

BeatWire - About the team

Team Photo

BeatWire consists of Richhiey Thomas and Oliver Watson.

Richhiey Richhiey

Oliver Oliver

Both of us are masters students studying Data and Knowledge Engineering at OVGU Magdeburg. At university, we both have learnt about MIR and deep learning. Richhiey had participated in the song contest previously whereas Oliver has completed a few projects in MIR in the last semester. We both were always friends and got together for the song contest to explore the possibilities that lie in this space.

Acknowledgments

Many thanks to the AILab at OVGU for helping us out with equipment, providing a workspace and guiding us in this space of AI and music. Also, thank you to family and friends for listening to various stages of our music making process and providing feedback!