Musimap scored a coup a few weeks ago when it announced that the composer, producer and entrepreneur Quincy Jones had invested in the Belgian start-up. Musimap is developing products based on emotional artificial intelligence to help better grasp what kind of music consumers like. But the technology can also be exploited in others areas such as advertising placements and vending recommendations. Emmanuel Legrand asked Musimap Managing Director, Patrick Zucchetta, how the legendary artists got hooked on AI.
How did you convince Quincy Jones to invest in Musimap?
Patrick Zucchetta: We approached Quincy Jones at the last Montreux festival in 2019 through one of board members, Vincent Favrat [an early investor in Quincy Jones's Qwest]. Quincy was interested in the technology and by AI. We told him that we were working on improving our product and were planning to have an investor's round. He said he was interested in receiving the investment deck. What decided him is when we ran his emotional profile based on a list of tracks he likes and the result was very good. He said that it reflected who he was. That's what decided him to join the business.
How much has he invested?
We don't comment on individual investments.
Overall, how much did you raise with this round?
About 600,000 euros, if you include the whole round from shareholders plus credit lines.
Quincy Jones isn't simply an investor, he also acts as an advisor. How does that translate day-to-day?
It's a minority investment but what is more important is that he believes in what Musimap does and is ready to assist. Musimap is still a small company. We are a start-up in Europe so sometimes it is difficult for us to be visible from large companies. The fact that he is promoting Musimap is important. The moment we announced his involvement, we got immediately requests of interest. We are now above the visibility line and I am sure that in the long term it will be a real benefit for Musimap. Our technology is working quite well and we are partnering with some large catalogue owners like Warner Music Group so with more visibility, we will get more companies looking at what we are doing.
You also opened the capital to new investors.
We welcomed Andreas Spechtler, who is an ex-VP of Dolby. I knew him from the past. He acquired assets from one company I was driving and we kept the discussion open. He joined us at the beginning of the year and is now serving as chairman.
We have several products. The company started three years ago with MusiMatch, a service of audio similarities. If a music sync service wants to use a track from a celebrity but it is too expensive and they want to find something similar but more affordable, MusiMatch will find it. That was the first level. Two years ago, we introduced the next stage, MusiMotion, an advanced tagging engine. The idea is to describe a track with moods, genre and also, of course the acoustical analysis made by Musimatch. This is the product and service used by catalogue owners like Warner Music Group, with whom we signed a license last year. We have tagged 1,5 million of Warner songs. We are also working with streamers like Qobuz in France, which has 3 million tracks enriched by MusiMotion. And on top of this, we released MusiMe, our psycho-emotional engine, which provides the links between tracks and individuals. You give 10 to 20 of your favourite tracks and will profile you according to industry standards. This is what we did with Quincy. Based on the result, you can integrate the profile into services and it can advise listeners to listen to that track instead of that tracks. And now we are working on a recommendation engine.
Could it be used by streaming services?
We are a B2B service, so if we take the example of streamers, and Qobuz is a good case, we deliver tracks enriched by MusiMotion but also deliver a recommendation service. By knowing what the audience is listening, we can build profiles and can advise on listening recommendations.
How do you monetise your services?
If we speak of MusiMatch and MusiMotion, the price is based on the size of the catalogue. We have a price per track model. For MusiMe, it is a mix between the number of tracks and users. That's the model we use for Qobuz. For this to work we need to have tracks that have been analysed and enriched. That's the first step. To give you an idea, after we delivered enriched tracks to Warner Music and they passed on the tracks to Amazon. They wanted to test the quality of tagging so they ran a test and compared our tagging with what their musicologists said and we were very high in similarities. That's a good test.
How do you keep up with the changes in technology?
We are constantly working to improve our systems, especially the psycho-analysis of the audience. As a user your mood can change during the day so the profile needs to be updated. Most recommendation systems are static. We are using AI but the first step is to make sure that it is as good as possible on a large quantity. Musimap came from scientists. We have one million tracks that have been analysed by humans and from this we have a model. Thanks to AI we can do it in a faster mode now. Analysing one million tracks required 10 years of work, while right now, with the existing infrastructure, we can manage 5 million tracks every month. We are already at the fourth generation of our system. Every nine month, we improve the AI model to release a new version, and it is more accurate.
Could your system work for A&R to predict successes?
Not yet but we are working on it. We are part of the FuturePulse project from the European Union, which includes the likes of Spotify and IRCAM. This is on our roadmap to release track success prediction.
What's your endgame?
We have two stages: First, expand into the music sector, and definitely tag more catalogues. We currently have five million tracks, and we'll be happy with 50 million or more. Then we plan to move outside the music sector. We are working with a company in the advertising sector to be able to use MusiMe to recommend advertising spots. We are working with a group of radio stations in Germany to help fine tune an engine selecting spots. And the next step would be going into e-commerce. We are looking at connected speakers and how your profile can be used to move products that are more appealing to you and, through a seamless process, recommend more products.
Are there limits to AI?
There are no real limits. But AI is never giving you a fully accurate picture. What we deliver is percentages, and the evolution is based on accuracy. Our target is to be more accurate and this is how clients will judge us.
No comments:
Post a Comment
Note: Only a member of this blog may post a comment.