April 27, 2019 weblog
Bach and Adele: Knock yourselves out on MuseNet
OpenAI is introducing a musical MuseNet, the music-generating AI that was in the news earlier this week. Some AI watchers were calling the music OpenAI just unveiled as amazing.
Victor Tangermann in Futurism said the tool "generates musical compositions, with multiple instruments and in wide-ranging styles from classical music to country music."
OpenAI said that the MuseNet, a deep neural network, can generate 4-minute musical compositions with 10 different instruments. It can combine styles too. The researchers taught it to produce convincing, "genre-bending" tunes.
Genre-bending? Like how? Imagine notes from Mozart sent forth to tromp around "Poker Face."
OpenAI researchers trained a deep neural network on huge data sets of jazz, pop and world music, said Futurism.
On the site, you can choose between a number of famous classical composers in order for the AI to come up with a composition in their style.
All it needs to do its thing is a starting point, as maybe Mozart's and Lady Gaga's "Poker Face." This prototype of a MuseNet-powered co-composer will be available through May 12 and then the team will chart a next step based on feedback. It allows a subset of MuseNet's options.
"We're excited to see how musicians and non-musicians alike will use MuseNet to create new compositions!"
Thankfully, Will Knight in MIT Technology Review is interested in the technical powers of delivering music but that does not mean he forgets the value of real music. "A powerful AI algorithm can dream up music that echoes Bach or The Beatles, but it isn't real creativity," said the article's subhead.
Knight took the question, is this really music? and posed it to Zach Lipton, an assistant professor at CMU and jazz musician. What did Lipton think about MuseNet's jazz improvisations? His reply: "It is uninteresting in precisely the same way as every generic 'we trained an LSTM to generate ____'. I don't think there is anything here that a musician should find interesting."
Knight's own observation is that music is not only rooted in culture, history, and language but has "remarkable capacity to surprise, shock, and inspire. The algorithms have some ways to go yet."
In the bigger picture, OpenAI's MuseNet is contributing to a musical AI scene that is fast evolving.
Kyle Wiggers in VentureBeat took a look at recent activities. "In March, Google released an algorithmic Google Doodle that let users create melodic homages to Bach. And late last year, Project Magenta, a Google Brain effort 'exploring the role of machine learning as a tool in the creative process,' presented Musical Transformer, a model capable of generating songs with recognizable repetition."
© 2019 Science X Network