Introducing Musixmatch AI: the new Music Intelligence
“(Because I’m happy), Clap along if you know what happiness is to you”
Analysing music has always been a challenge. “Play a happy song”. Happy to run to? Or happy to enjoy it with my kids? Happy for a Friday night with friends?
Today, we are announcing a big stride that we are taking in order to answer this question.
Trying to find “a happy song”
Historically, all the music services have been using BPM (beats per minute) analysis to classify and recommend tracks. This has been a good start, but a fairly basic one. The tempo of the music doesn’t necessarily relate to the mood of the song. An example: “Killing me Softly” by the Fugees, Snoop Dogg’s “Drop it Like It’s Hot” and Radiohead’s “Creep” have the exact same BPM. To get the differences try reading the lyrics for the 3 different tracks.
Another music classification system started way before: the genre.
Another bug in the music classification as genres do not dictate the mood of a song: a pop song can be relatively sad — take The Killers “All These Things That I’ve Done”, for example. Just because a track is classified as “rap” doesn’t mean it is an angry song. For instance, “Good Day” from Nappy Roots, is clearly a happy rap tune.
Music is one of the most contextual personal experiences that exist. And lyrics are the context needed in AI to analyze music sentiment. That’s why today we are announcing Musixmatch AI — the first publicly available large-scale lyrics datasets for Music Information Retrieval.
The secret emotion of lyrics
Our goal is to build a system that delivers sentiment based on the content of the lyrics.
Yep Lyrics. The words of a song. The expression of the songs.
The language of the Music.
To achieve this, we are analyzing more than 14 million lyrics from our catalog , in more than 50 languages, and matching each one of them with a tone. We’re building this in order to classify emotions in the lyrics contents to know if they include things like anger, fear, joy or sadness, etc
Further down the line, Musixmatch AI will detect and interpret sentiment, social tendencies, language styles and meanings in song lyrics, helping to analyze the messages that artists express in their songs.
We’re doing this through Machine Learning and Deep Learning algorithms, trained by our immense community of 30 mln users.
We know what the lyrics are saying, but do we know what they mean to say?
Musixmatch AI uses the state of the art word embedding and text analysis algorithms to capture similarities among different concepts expressed in lyrics. By extracting entities, such as people, places, events and concepts, we are building a linguistic model to see tendencies that will tell us the degree of openness, conscientiousness, extroversion, agreeableness, and emotional range of each lyrics.
We’re building a new experience of music discovery with an innovative recommendation system based on lyrics and music languages.
Musixmatch AI will provide matches to songs that have concepts, insights and sentiment in common, all through the lyrics. All in a single entry point.
Accessing Musixmatch AI
We will be releasing several tools in the coming months that will provide a wide range of applications for lyrics text analytics and lyrics insights to empower organizations, artists, music services, machine learning researchers, and independent developers.
We’ll have frequent updates on our Musixmatch AI page.
If you have questions we’ll have the answers.