Participants 2022
About the TEAM
I’m Rui Guo from China, a fourth year doctoral student in music technology from University of Sussex, Brighton, UK. Currently I’m a visiting doctoral student in University of Liverpool, Liverpool, UK. My research is AI for controllable symbolic music generation.
I play piano from 7 year old and a little guitar from 18 year old. The genre I am familiar is mostly pop and classical. The deep learning model I make for music infilling is also targeted at pop music composition.
I’m based in Liverpool from last year, and it is good to have something for the contest from this music vibrant city. My music taste is mainly east asia, and I like to listen to light songs to soothe from work.
The team name “snagon” is a combination between “snake” which is my Chinese zodiac, and dragon which represent the Chinese identity. The one of team photo was generated by AI from this combination, and another photo was generated by “birthday trip”.
About the SONG
One of my favourate song is “heart beat” https://youtu.be/J8ftiX69Q1g
My AI tool can have variations of an input song, and I used that song as template.
I used an online lyrics generator with the first prompt “On Sunday I get up early, take the gift from far away”, which is to describe a birthday trip. The final lyrics changed much to fit to the song melody. All the lyrics are generated from the online service although, and I piece them together and adjusted them.
I don’t have a specific theme when I start the process. The generated new melody is lighthearted and that made me to make this song light and joyful.
About the HUMAN-AI PROCESS
I use my own pop song composition tool based on AI, which is an plugin in Ableton live. It can rewrite the existing music, e.g. a new melody given the bass and harmony, or a new bass in bar 5 given the rest of the music. The best thing for that tool is it can control the texture of music, including density, polyphony and occupation, for each track in each bar, or the average level of 16 bars. It made the generation process much easier with controls, and I select the version I like.
I first generate a new melody, and then bass and harmony. Then I adjusted each bar for those tracks with the new versions from the AI tool, and select the one I like most. By those controls, I can apply my music idea to control parameters, and let AI fulfill the result.
After generation of those three tracks, I generated a guitar track as another accompaniment track along with piano. Those two accompaniment tracks are panned to different channels.
The vocal are synthesised by a free version of piaprostudio for vocal synthesis. That free version can synthesise English singing sounds, in a voice called “miku_v4_demo”.
The generated song still follows the template song structure and harmony progression, but it can be hardly discriminated. I deliberately use the original songs ending as the ending of my song.
I changed some of the melody to fit to the lyrics. I added the drum track myself. For bass, piano and guitar, they are largely intact from the AI generation. I paid a music producer for mixing/mastering, and we had 6 hour live session for this purpose. I also made a music video by AI, thanks to the code from https://github.com/yk/clip_music_video
Lyrics
On Sunday it’s a little gloomy, I get up early it's your birthday.
Clean my best Sunday shoes, take my eyes from tired and dress always in blue.
I have to attend, this is the day and get on a train.
I'll take you out with a bang and take the gift from far away.
Waiting for you now, night and day.
Oh please tell me we'll be safe.
I don't run away, light up the day
I promise you I'll wait as it takes and I know I'll leave someday.
I know something I should say.
On your birthday.
On your birthday.
On your birthday.