The label Snafu Records update on the AI to unearth the stars of tomorrow


The label Snafu Records update on the AI to unearth the stars of tomorrow

The start-up Swedish has developed its own technology of machine learning to detect on the Internet the singers anonymous to high potential. With success so far.

Founded in march 2018 in Stockholm, Snafu Records is a label of independent music is not quite like the other. Focusing 100% on the Internet side distribution, it is based on an AI for spotting potential future stars. “We are a record label of the 21st century that uses the technologies of the 21st century,” says Ankit Desai, founder and CEO of Snafu Records. Officially launched in early February, Snafu Records already counts more than 20 artists in the portfolio. Among them, there is the jazz singer costa rican Mishcatt, from which the song Fades Away has been listened to 5 million times during the five weeks following its release, or even the pop-duo american Joan who has passed the bar 55 million streams.

Ankit Desai, founder and CEO Snafu Records (right), with Carl Falk, the creative director of the label. © Snafu Records

Freed from the constraints of time related to reading “manual” of songs, Snafu Records, which is limited in time to 20 employees, can cast a wide net. Each week, its AI is sifting between 170,000 and 190,000 pieces on platforms such as Soundcloud, Spotify or Youtube, and selects 200 to 500 of them. “5 million songs are broadcast each year on the web. It is impossible to listen to everything even if you deploy thousands of people. The artificial intelligence solves the equation,” says Ankit Desai.

The AI of the start-up is well-oiled. Via learning algorithms oriented time series, it considers first the velocity with which the pieces could grow in terms of audience based on their history (number of sheets, of lives…), and then deducts from their revenue-generating potential. Models of analysis of feeling then enter in the dance to assess the level of engagement of internet users vis-à-vis songs. Finally, the IA examines the musical structures and compares them to those of the 200 pieces of the most popular on Spotify.

Related Post:  Charli D'amelio : age, sister, partner, salary... everything we know about the star of TikTok

An office in Los Angeles

“Based on these criteria, we are able to identify, for example, a colombian singer unknown in the world dream to hear, and we will be able to emerge. Knowing of course that some event, typically a transition to the television will accelerate the growth estimated in the beginning,” admits Ankit Desai. For the time being, Snafu Records is based mainly on models of machine learning, statistics, type regression and support vector machine. But the company also makes use of neural networks (or deep learning), in particular for sentiment analysis.

With the ambition to play in the big league, the Swedish company opened its first overseas office in Los Angeles. To finance its growth, she completed two rounds totalling $ 3 million. It may also depend on its creative director, Carl Falk, famous music producer and song-writer in Swedish, who wrote for Madonna, Nicki Minaj or even the boy band One Direction. Like a label to be traditional, Snafu Records is paid by taking a margin on the revenue generated from the sales.

“Music is a form of universal communication, our algorithms are agnostic in terms of languages”

The company is said to be the first label to have put the AI at the heart of its activity. Of course, there are other start-up developing technologies AI equivalent, such Instrumenal or Chartmetric. But it is the vendors and not the label.

“For the moment, our algorithms perform better in terms of detection of talent that, in terms of sales projections, because the calculation depends on a number of factors significantly more important. But our models are continually and we already have two artists (mentioned above, ed) on a score who have attained an international stature in the space of 12 months”, argues Ankit Desai. “Knowing that the success rate of major labels amounted to 5.7%, it is clear that our approach-oriented data is much more efficient.”

Related Post:  The friendship tumultuous that lia George Michael to Elton John

Predict the next tube

The priority of Snafu Records in terms of R&D ? “We are now looking to predict that a song will become a tube even before it has been broadcast, without having to analyze the way in which it is consumed or hosted on the web, based only on the history of the artist and song structure,” says Ankit Desai. To meet this challenge, the young shoot put on the deep learning.

Other works referred to : the international development. “Music is a form of universal communication, our algorithms are agnostic in terms of language. Thus, they can detect talent, regardless of the country,” says Ankit Desai. “Still, the final decision requires to master the local culture.”

Of european origin and implanted in the United States, Snafu Records already covers the cultures, anglo-saxon, and several european countries, including Sweden but also to France, where its CEO has lived for several years. On this basis, the company intends to double the number of its contracts by the end of the year, rising to 40 artists.


Please enter your comment!
Please enter your name here