If Frank Zappa’s Endless Guitar Solos Somehow leave your earbuds to crave more, music-hackers Dadabot are here to fulfill your listening desires.
The team of CJ Carr and Zack Zukowski recently used a Recurrent Neural Network (RNN) to generate a boundless bass solo Live-streaming on YouTube.
Do tShowered his neural network on a two-hour improvisational bass solo by composer Adam Neely.
“My strategy was to use the same bass, same pick, same tone, improvised baseline on the 85 beats per minute drum groove and keep everything in the e-minor, so the end result wouldn’t be too colorful.” Explained on youtube.
Dadabotts ran the recording through SampleRNN, an audio generation model that analyzes a piece of music to infer the sounds that come forward.
His initial effort produced too much noise, a “slow and sappy” bass pattern, and a “soulful” timebra, so he fine-tuned the dataset and dropped the sampling rate. This eliminated noise and helped to learn RNN patterns that were twice as long.
Dadabotts then curated the data for mostly rapid drop-offs, as the sample RRNN works best with faster speeds.
He also experimented with different temperature values, a Settings that control the randomness of the output:
Sometimes the high temperature rings like two bass at the same time. Like it’s some sort of … shadow bass …
The resulting single can be grating, but surprisingly mellow as well.
The project shows how AI’s role in music can move from a gimmick to a powerful creative tool. Alternatively, it can produce an endless stream of original music with relatively little human effort.
Dadabot is exploring a range of possibilities. They have already produced a diverse discography AI generated tunes, And an as-yet-unreleased fusion of funk and progressive rock that is full of 13 albums.
It would probably be a little painful to sit in one sitting, but I’m sure there would be some musical gems – or at least some sounds you think.
Until then, it’s good to see Very funny The art of bass guitar is in the headlines.
HT – Engadget
Published December 23, 2020 – 18:09 UTC