FutureFive New Zealand - Consumer technology news & reviews from the future
Story image
Startup launches AI-powered music curation service
Fri, 12th Jul 2019
FYI, this story is more than a year old

A Singapore-based start-up, Musiio, has launched a new music curation service powered by AI technology.

Dubbed DIY AI, the service allows users to input a link to a song of their choosing, and within approximately 10 seconds Musiio will generate at least eight custom tags, including genre, BPM, key, and mood.

Musiio has created an AI model that uses audio fingerprinting to surface, tag and sort tracks on a large scale. The platform can also generate playlists based on these features.

It has been trained on music from around the world and can recognise many genres and styles, especially East Asian and Pacific styles and languages.

For music that falls within its largest training sets, such as Western pop, its accuracy hits 99%, according to Musiio.

The DIY AI demo, recently launched by Musiio, enables anyone to try the technology and see how it better enables search, discovery, and curation, for everything from sync to ambient musical applications.

Musiio co-founder Hazel Savage says, “Smartphones and home studios have brought about the democratisation of audio production, but one of the challenges for the music industry now is to figure out how to handle this volume.

“It takes one person 83 days non-stop to listen to 40,000 new songs. Our AI can perform the task in under four hours.

“As music is created and released at a rate that is one million times greater than it was ten years ago, AI can solve problems that humans simply can't,” she says.

Musiio was a member of the Entrepreneur First incubator and a Midemlab finalist. It began as a tool for search - specifically for finding the right track in huge catalogs.

The fingerprinting technology, which turns the audio file into mathematical visuals, allows someone to drop a file in and find 10-20 other files with similar sonic features.

As Musiio evolved, the team discovered that tags were in high demand. They taught their models to compare audio features, find similarities, and connect them with appropriate, customisable tags.

“We didn't build it around assumptions of 'this is what music is'. You don't need data; you just need to train the model to match patterns,” says Savage.

She says, “We're focussed on the curation of music using AI, not on generation or other aspects of the creation and listening process.

“It's important because in our world, AI for music is not scary: it contains no sentience; it definitely isn't a robot trying to steal your job, quite the opposite in fact.

“We aim to help artists be found and labels and streaming services to deliver a better, more personalised product.