Traditionally, success in the music industry has always been closely associated with touring, sold CDs, and charts. With the rapid digitization of our world, the music industry has moved to a new way of measuring success — data. With so much information about music at hand, data science consulting companies like Iflexion can build specialized solutions to identify which songs a particular person will like, predict the next big music star, and craft songs to suit a very specific target audience.
How Data Shapes Sound
As much as underground communities, genre gatekeepers, music critics and art enthusiasts would love it to change, the music industry has been heavily commercialized. Commercial artists’ task here is to create music that will satisfy big audiences and make a profit. When data used wisely, artists can write songs that will more than likely appeal to a particular audience.
One of the most prominent data advocates in the music industry, Ankit Desai, has once looked at streaming statistics of Swedish artist Tove Lo. He noticed that one of her songs was particularly popular among EDM fans. Desai advised to capitalize on this opportunity to win a bigger market, and two months later, Love To released a song featuring EDM artist Alesso. The song went platinum in a number of countries and made it to the top of The Billboard’s US Dance Club Songs chart.
How Data Made Spotify Superior
With 286 million active users and a nearly 40% share of the global music streaming market, data is a critical factor of Spotify’s worldwide success.
What sets Spotify apart from competitors is its powerful recommendation service. Each Monday, every user receives a customized ‘Discover Weekly’ playlist that is comprised of 30 songs specifically selected for every user. Such an extreme level of personalization is possible because Spotify acquired at least six music recommendation and machine learning-related companies including Niland, Sonalytic, Seed Scientific, and The Echo Nest.
Currently, Spotify uses a combination of these three recommendation models:
– Collaborative modeling. Spotify’s machine learning model constantly analyzes what type of music you currently like by evaluating your actions towards particular songs. For example, the algorithm takes into consideration which songs you’ve played on repeat, added to the playlist, etc. Then, Spotify compares your music preferences to other users, finds those with similar tastes, and recommends songs they like to you.
– Natural Language Processing. After scanning a track’s metadata (artist name, song title, etc.) Spotify’s NLP model scans thousands of articles, forums, blog posts, and discussions about an album or a song on the internet. The algorithm analyzes what language people use to describe the song and matches them with other songs that are discussed in a similar manner.
– Convolutional Neural Networks. Spotify uses a CNN-based model to analyze raw audio data regarding the song’s BPM, musical key, loudness, and other parameters. Spotify then finds songs with similar parameters and recommends it to you. This model has proven to be exceptionally effective for discovering quality music that is yet to be recognized by the masses.
How AI Lowers the Music Industry’s Entry Barrier
A great song is a combination of one’s creative spark and others’ technical knowledge. You would be surprised how much it takes to transform a dry and lifeless recording into what you hear on the radio or a streaming service. The process of optimizing a track for an adequate listening experience is called mastering. For many up and coming artists, everything related to audio processing is the biggest roadblock on their way to releasing music, as professional mixing and mastering services usually cost more than they can afford.
1. Machine Learning Concepts Every Data Scientist Should Know
2. AI for CFD: byteLAKE’s approach (part3)
3. AI Fail: To Popularize and Scale Chatbots, We Need Better Data
4. Top 5 Jupyter Widgets to boost your productivity!
This is where AI comes into play. For example, LANDR is an ML-powered online service that can master a track in a few minutes. LANDR’s algorithm took thousands of professionally mastered songs as a blueprint and now matches audio qualities of those songs to the uploaded ones.
Similarly, Soundcloud, one of the most popular free online music streaming platforms among independent artists, has also introduced an online mastering tool powered by ML. Currently, there are many AI-assisted audio plugins and tools that help less tech-savvy creators produce music of high sound quality.
How Data Helps Discover New Talent
In 2018, Warner Music Group acquired Sodatone, a service that feeds streaming, social media, and touring data into ML algorithms to identify which artists have the most potential to become successful in the future. This year, Amazon patented its own technology that can predict the future popularity of various media content including music, books, and films.
Hitlab, a Canadian digital media and AI company, aims to be the major tool for AI-driven A&R. Music Digital Nuance Analysis (DNA) is a patented tool that helps break down any song into 83 attributes. The tool can analyze the most popular songs in any region and then compare their attributes to any newly released song to identify the ‘hit’ potential. This can become a secret weapon of modern-day producers, songwriters, labels, and publishers as now they can tailor their sound to appeal to a specific target audience.
Will A&R professionals become obsolete? In short, definitely not. As in most other AI use cases, the technology here will become more of an assistant. With 20,000 songs uploaded to Spotify every day, the scouting job becomes increasingly difficult. Such tools will only help narrow those thousands of songs to a hundred, significantly easing the A&R job.