A new neural network produced by OpenAI takes the concept of manufactured pop to whole new levels.
The artificial intelligence (AI) system is called Jukebox and it can create music from scratch, complete with lyrics and rudimentary vocals, in a range of genres and styles.
This isn’t the first time that AI has composed music, though previous efforts have often centred on classical and ambient electronic music.
While it is more complex than pop music and other contemporary genres, classical also tends to follow strict mathematical rules.
Adding lyrics and vocals also adds a whole new challenge.
OpenAI, a non-profit research organisation backed by Tesla chief Elon Musk among others, had previously demonstrated its own MuseNet.
This was described as ‘a deep neural network that can generate 4-minute musical compositions with 10 different instruments and can combine styles from country to Mozart to the Beatles’.
PG Music’s Band in A Box and Google’s Musical Transformer have also been able to create melodies if not fully formed songs.
Jukebox can create songs in a range of styles
Jukebox takes things several steps further with a capacity to create music and lyrics in pretty much any popular or classic style, including country, hip-hop and heavy metal.
OpenAI used a massive dataset of around 1.2m songs to train its AI.
Each song also contains masses of information, with a typical four-minute CD-quality recording having over 10m timesteps.
In order to make the processing of all this information more manageable, researchers used an automatic encoder to compress the data and remove irrelevant information.
Even then, it takes around nine hours to render a single minute of audio.
Despite all this, the results are not perfect, as OpenAI itself admits.
The organisation said: ‘While Jukebox represents a step forward in musical quality, coherence, length of audio sample, and ability to condition on artist, genre, and lyrics, there is a significant gap between these generations and human-created music.’
One area for improvement is the ability to create repeated choruses.
Recordings can also include noises that sound strange to the human ear, and the system is as yet trained only for Western music and lyrics in English.
It does represent a big step forward in one of AI’s biggest challenges, however: imitating human creativity.
Today’s news was brought to you by TD SYNNEX – the UK’s number one solutions distributor.
