Artificial intelligence A&R tools keep finding new hit artists. This one actually ‘listens’ to music.

Hazel Savage

To illustrate the importance and potential of AI-based A&R, Musiio co-founder and CEO Hazel Savage refers to the origin story of one of the biggest breakthrough UK artists in recent years, Lewis Capaldi.

Her point isn’t that the self-effacing Scottish superstar was discovered via data, because he wasn’t. Her point is, What if? She refers to him as the one who got through, as opposed to the thousands who got away.

She says: “With millions of creators posting new music to UGC sites around the world, the job of listening to it all has become impossible.

“Lewis Capaldi was discovered by manager Ryan Walter, who famously spent six months listening to every new artist on SoundCloud he could find. He would listen for seven hours a day, with up to 500 tabs open at a time, listening to 10 seconds of each track.

“Capaldi’s rise to fame is a fantastic tale, one which dreams are made of, but the truth is 99.99% of the work that goes into finding artists is never seen or recognized. It can be a brutal process and if Capaldi’s story teaches us anything, it’s that there is talent out there that could quite easily go unnoticed and not listened to.”

To try and help fix that undoubted problem, Musiio has created a new piece of technology that it calls a Hit Potential Algorithm, which it claims will not only be able to classify and categorize new music, but also accurately measure hit potential, and isolate those tracks most likely to properly blow-up – no matter where or who they are from.

Here, Savage tells MBW about the new technology’s secret sauce and how it could radically streamline the A&R process…


Can you start by giving us a refresher on what Musiio is and the technologies you have introduced so far?

Musiio is a Music Tech start-up based in Singapore, but with a British and Swedish co-founding team. I have previously worked at Shazam, Pandora and Universal Music.

“We use AI to ‘listen’ to music.”

We use AI to ‘listen’ to music. What we mean by that is we transform audio files into something a computer can read. We do this to enable auto-tagging and audio reference search. Our core tech is Artificial Intelligence (aka Machine Learning).


Your latest move is an AI-based Hit Potential Algorithm. What prompted you to explore that space?

I have always believed this is the most powerful way to use our tech, the challenge is that it’s the hardest one to prove or sell to others. Music is more valuable the higher up the food chain you get, so to be ‘discovering’ music is the best way to create value.


What, specifically, does this new technology do – and how does it do it?

Our technology processes whole audio files. We ‘extract’ features from the audio files and we train the AI to look for patterns.

You can use this process to replicate anything the human ear might be able to recognize, such as the BPM and genre, right down to how much of a hit something might be.

Our new technology is looking for hits, virality, talent and melodic familiarity as well as a bunch of other factors that make up great music.


You’ve said that ‘volume is the new frontier in music’, what do you mean by that – and how does the new technology relate to that belief?

When I used to work at HMV in 2004 I put the new CDs on the shelf for sale Monday morning.

In those days, all new music came out on the same day and an average week might bring five or 10 CD singles. Streaming services get 40k tracks a day uploaded to them, so what used to come out in one year now comes out many-fold every day; great for artists, but it does pose challenges.

“AI lets you automate some parts of the [A&R] processes. Ideally you automate the most tedious and unmanageable parts.”

That’s the volume frontier, and even more so because everyone has access to the same audio, so what you can do when an unmanageable volume of music becomes the new business models.

AI lets you automate some parts of the processes. Ideally you automate the most tedious and unmanageable parts.


How have you tested the technology and what have the results been?

Our more readily available technology, such as AI tagging and AI search are being used widely (Epidemic Sound uses our tags for example), and our APIs are for sale anytime.

The Hit Potential Algorithm is a brand new product that has been in R&D for the last three years. Our testing is in-house, but just like our other products which are 90-99.9% accurate, we believe this new product is too.

We never release products that don’t do what we claim. Our reputation is important to us, and it’s why we hire skilled AI developers and everything is built in-house.

“Imagine only having to search through 20 songs instead of 10,000 to find bangers.”

We took 10 viral hits from YouTube, hid them in a database of 10k tracks from the Free Music Archive (creative commons database of music with varying quality) and asked the AI to find the top 20 hits. It found 9 out of 10 viral hits in the top 20.

Imagine only having to search through 20 songs instead of 10,000 to find bangers.


Where is it up to in terms of readiness to use – and where will it be applied, by which companies?

The core technology is working and I am excited to announce the technical capability. How and where we might apply it is in stealth mode.


Aren’t there similar technologies already active in the marketplace – what makes Musiio’s offering unique?

There have been many approaches. ‘AI for Music’ is a very generic term.

Sodatone, which is a great company that was bought by Warner, uses data aggregated from open API sources (such as Spotify) and some private label sources (Warner‘s own data), to look at play counts, likes, followers and see what is trending.

“A track can have zero plays, no release date, unsigned UGC and no social media, and we can still tell you if it could be a hit.”

This is not a bad approach and you can learn a lot from this data, but you still need access to the data and it still needs to ‘trend’ ‘hit’, to some extent, for it to get flagged in a data science AI algorithm.

Others have claimed to ‘use AI’ to identify hits or do A&R in a similar way, aggregate large volumes of data and spot trends. It has its place, but to the best of my knowledge Musiio is the only company working on the audio file itself. This means a track can have zero plays, no release date, unsigned UGC and no social media, and we can still tell you if it could be a hit.


What might it mean for artists and managers?

I hope it means more discovery for hidden artists. For every Lewis Capaldi there could be hundreds of undiscovered talents, we hope this technology will help discover them.


What do you say to traditionalists who baulk at the emergence of AI/algorithmic A&R and who argue that the most important tools will always be ‘good ears’ and ‘gut instinct’?

I think good ears and gut instinct are important; people are important. AI should free them up to do what they do best.

Let the AI process 40k tracks a day and pass an A&R the top 20, then let a person make the final analysis. The alternative is someone listening to 40k tracks a day, which is not humanly possible. It’s about increasing the odds, not replacing people.

Also, let’s be honest, true ‘tech doubters’ are starting to sound like the folks who hated synths in the 80s [laughs].Music Business Worldwide

Related Posts