One of the big talking points around AI music is scale: the sheer amount of music these models are able to generate once they’re up and running.
Here’s a new data point on that: startup Mubert says its AI has now generated more than 100m tracks, including more than 56m created by its users.
As the company pointed out in its announcement, that’s roughly the same number of tracks as the average music-streaming service has in its entire catalogue.
Mubert has been going since 2017, and its AI is put to work over several different services. Mubert Play is its consumer app, providing functional music to help people relax, focus and sleep.
Mubert Render, meanwhile, is a tool for people to create new tracks using its AI, which Mubert says is used by more than 100,000 monthly active users.
The company also has an API for its B2B customers, and has worked with partners including streaming service Anghami and virtual world Sensorium Galaxy in recent years. Last October, it also released a demo interface for its ‘text-to-music’ system.
The company is also keen to stress that its AI hasn’t been trained on commercial music without permission or payments. Mubert says it has built up a library of 2.5m ‘proprietary sounds’ to train its model, including paying musicians for contributing samples.
Like several other startups in its field, Mubert is setting its sights on the creator economy – people making videos for online / social-media usage who need soundtracks – as a key source of business.
“It is impossible to imagine streams, podcasts, and shows without music, and Mubert allows for the generation of an unlimited amount of music of any duration and any genre, tailored to the needs,” said co-CEO Alex Mubert (pictured above) in a statement accompanying the 100m milestone.
His company is not the only startup to publicly say how many tracks their AI has been used to make. Boomy, which launched in 2019, publishes this metric on its homepage, and it currently stands at 16.3m songs.
One caveat for this kind of statistic: it only shows how many tracks have been created using an AI, but not how many of them were actually good.
The traditional way services like Mubert Render and Boomy work is that you create a song using the AI, listen to it, and if you don’t like it, discard it immediately and get it to try again.
In the future, a more useful metric might be the ‘hit rate’ of these AIs, in terms of the percentage of tracks created with them that have been saved, shared and/or used rather than discarded.
Tools: platforms to help you reach new audiences
Tools :: Wyng
Through Music Ally’s internal marketing campaign tracking, we’ve recently discovered an interesting website by the…