July 17, 2017:Startup Muru Music sees billion-dollar potential in ‘first AI DJ brain’

Startup Muru Music sees billion-dollar potential in ‘first AI DJ brain’

“I think we have the opportunity to be a billion-dollar company if we attack all verticals…”

For a startup with just two full-time employees and AU$500k (around $386k) in funding so far, setting sights on a 10-digit valuation is certainly an arresting claim.

Muru Music founder and chief musicologist Nicc Johnson, however, feels that by focusing on metadata and discovery issues for everyone in the digital-music chain – from publishers and labels to digital services and broadcasters – his startup can unlock some enormous opportunities.

The company describes itself as “the first AI DJ brain” and sprang out of Johnson’s past as both a DJ and a music supervisor, plugging the gaps that he felt were compromising the discovery algorithms on digital services.

Dutch by birth, he grew up in Ibiza and cut his teeth as a DJ on the island during the summer season, as Europe started to embrace electronic music. In 2011, he moved to Australia and three years ago developed what was to become Muru – taking its name from the Aboriginal word for “journey”.

Johnson was, at the time, consulting to the hospitality industry and providing playlists for hotels and bars but felt they could be markedly improved.

“I had an idea for a system that I wanted to build to improve my workflow and provide a better service to the clients,” he tells Music Ally. An initial investor and early angels came on board, with the company raising AU$500k to date, with another funding round due to close soon.

Johnson and ‘chief engineering wizard’ Vijay Santhanam are the only full-time staff members in the Australian office, although they do hire designers and developers on an ad hoc basis.

“When I started researching this, I wanted to create a tool to create playlists for myself – that was the main thing,” he says. “When I started looking at streaming services and what they offered, I realised that they had no search; like if you wanted, say, 20 deep house tracks, that wasn’t an option. So I started to think about how I could build that.”

He continues, “The aim was that, if I needed to create a 12-hour playlist, how I would get the content to fill that playlist? I found there was no real solution for that. When I investigated further, I discovered none of these services had any classification for music. The classification was all over the place.”

Nicc Johnson Muru Music

The streaming services have been trying to tackle this challenge. Pandora built its ‘music genome’ for example, while Spotify acquired The Echo Nest. Johnson thinks they have much more work to do, however.

“They had genre tags, but not necessarily classification. For example, Rihanna can have reggae, hip-hop, pop, funk and dance-pop as genres for her sound; whereas when I talk about specific classifications and take one particular Rihanna track, what is it? They don’t do that. They say it’s 60% this, 35% this and 5% this. That is great for certain things, but for a recommendation engine that is still very tricky,” he says.

I used The Echo Nest before it was acquired to do the tests. That’s when I realised their classification was horrible. I was using the Echo Nest data and they had a lot of really cool tools but, for what I needed to do, I couldn’t do it with their data or algorithms.”

Johnson was also linking into Last.fm and SoundCloud but felt they all meant he and Muru were coming up short. “We found that on all of these services, 90% of their tags didn’t influence recommendations,” he says. “We use 66 genre classifications and we provide better recommendations because of that.”

How does Muru Music classify “better” recommendations? That’s a qualitative call, surely?

“When it comes to recommendation systems, there is a grand truth to what a genre is,” he proposes. “Club music has a certain structure and within that deep house and disco have different qualities. We have identified the qualities of each individual genre and, based on that, we have trained a machine to learn what they are so they can find more like them,” he says.

It’s a machine-learning algorithm, so it’s deep learning and artificial intelligence combined. The secret sauce of our system is that we did not do acoustic analysis, which is what everyone else has done. We have built a very complex rule set around individual genres.”

Johnson says that Muru Music’s system looks at tempo, energy, year of release, popularity and so forth just like other systems do; but it also looks at labels associated with the music.

“We have the basic metadata for the songs that every label and every streaming service has – and we then layer that with a very complex rule set for each individual genre to identify the genre and then build a database around that,” he says.

Muru Music

Johnson’s grand ambition for the company is to become the industry standard that all copyright owners and services map onto their existing metadata to supercharge their recommendations.

In terms of the genre tags coming with the metadata, we actually want to be the foundation for all music,” he says. For now, the company’s technology can be seen at work in its free iPhone playlist-building app.

In an industry sense, though, the model is around SaaS (software as a service) where Muru Music will license its API to everyone in the digital music ecosystem.

“It is not an easy thing [to get the industry to move as one],” he accepts. “That is why we have to demonstrate through our technology and services why it actually works.”

Johnson adds that Muru Music’s system can classify songs at scale, which he hopes will be useful for digital services having to deal with a weekly glut of new releases ingested into their catalogues, as well as processing the tracks already in there.

Spotify and Apple Music are receiving 20,000 new songs a day and are only classifying a tiny percentage of those,” he claims. “So there is a lot of music that gets ingested into the system but is not included in the recommendation or discovery [algorithms]. We are solving that direct problem. Our system can classify those 20,000 songs [clicks fingers] every day.”

“We are the only ones that have classifications in this way and can prove to them that it is more accurate than anything they have.”

Muru Music may have its roots in the dance world, which is an area that Johnson believes has been under-served by the recommendation algorithms of the big streaming services. “A playlist of deep house might feature David Guetta which, as a DJ, is a sin!” he says. “We found the same with jazz and classical. We wanted to make sure that it worked for all genres.”

For now, Muru Music’s system covers 66 genres of music, all from the west. The next phase will be to apply its technology to music from Asia: for example, being able to distinguish between K-Pop, J-Pop and C-Pop from Korea, Japan and China respectively.

Is there an exit strategy here? Is Muru Music hoping to position itself, much like The Echo Nest, as a tempting acquisition by Spotify, Apple or one of their rivals?

“It all depends on the offer,” says Johnson, suitably coyly. “But the only reason I would look it is if they would provide the resources to get this to where I think it can be.”

Muru Music

As an independent startup, Johnson wants Muru Music to be as service-agnostic as it is genre-agnostic, although as The Echo Nest showed, an acquisition by a major streaming service can change that situation quickly.

“I want this classification to become an industry standard as I believe that it can change the entire industry, in the sense that I think all recommendation and all discovery – all music experiences on streaming – start with one thing, which is the classification,” says Johnson, although he is already eyeing other parts of the music industry.

I think we can disrupt various verticals – such as music licensing and sync, AI voice control and recommendations at scale. The other verticals are where I see the big opportunities for us,” he says. Which is when he voices his billion-dollar ambition.

“One vertical on its own? No. Attacking all the verticals we have in mind – and that includes hospitality? Yes.”

Eamonn Forde
READ MORE: Analysis News Startups
TAGS:
One response
  • rahul says:

    others could implement this as well. segmenting songs on a multi and micro level geners is no big deal and this geeky feature may be just too granular for music lovers….may be DJs will love to discover the recommended songs to copy bits and pieces and do dj mash-up from genre match-based recommendation but there are several apps to do that as well….will be interesting to see how muru will unfold as their features can be easily and readily copied by some of leading players…so where is the catch?

Leave a Reply

(All fields required)