Participants 2022

Y7

AI Song Contest 2022 / participants

TEAM / Y7
SONG / baisline_2.0
TEAM MEMBERS /
Y7

About the TEAM

Y7 is an audio-visual artist based in Stockport, UK., specialising in artificial intelligence. They have worked as a sound and video artist in a contemporary art context; as an ai specialist on music videos and tour visuals, and have recently begun to release their own music. Research into audio ai is at the forefront of their practice.

About the SONG

baisline_2.0 is the second instantiation of an ongoing research project into artificially generated UK bassline music (also known as niche bassline, UK bass or speed garage).

baisline_1.0 (https://youtu.be/XG5CJ7Ve_8Q) was constructed of small audio samples generated from one type of neural network, and was fairly low quality. 

baisline_2.0 however, is a mixture of 3 different types of raw audio neural networks (Finetuned Jukebox Models, SampleRNN & RAVE), all of which are trained on 3 ½ — 5 ½  hours of 44.1khz (standard quality) audio. From these models, over an hour's worth of audio was generated and whittled down by a human into one track. 

Like its predecessor, *everything* you hear in 2.0 has been generated by an AI, and very little has been done to modulate the sounds through effects or EQing. 

About the HUMAN-AI PROCESS

Of the raw audio that was generated by the various neural networks (which had a running time of over an hour), there were amazing parts that sounded like sections of finished songs. The problem with this is that you can’t separate the instrumentation (drums, bass, vocals) found in an audio file in order to flesh out further sections to create a complete song. 

In the end, the main feature I worked around was the regular vocal hits and short vocal melodies that appear at the end of some bars. These were all in the same key / scale and were flexible enough to allow me to pull together a song that had ebbs and flows and distinct sections. I then pieced together samples varying in length from half a second to 10 seconds to make the finished track.


When working with AI audio to construct a track, the role of the human is somewhere between a gardener, a ham radio enthusiast and a metal detector. During the training process you try and tend to a growing form of intelligence, attempting to improve the quality of its output by altering certain conditions (or pre-conditions). Once it has reached the stage where it can output things deemed of sufficient quality, it is then a case of scanning through noise to find salvageable artefacts.

Check out the other
songs of 2022