Tech understands lyrics to choose songs for entertainment industry
Finding the right song for the right moment can be a needle-in-a-haystack job for TV and movie execs.
A track that just about does the job isn’t going to cut it in the next Netflix mini-series. So they’re faced with trawling thousands of songs to find the perfect one.
Current technology lightens the load a little, tagging tracks by genre – but it isn’t very helpful when they’re looking for something very specific, like complex lyrics, a faster beat, or a song about a particular topic.
An AI developed in Israel takes a different approach. It goes through the catalog of record labels, which has thousands of songs that have never seen the light of day, and finds the perfect track in minutes, rather than in days or weeks.
It is uniquely able to analyze song lyrics and determine its theme, like success, a breakup, or violence. And it can identify what musical influences – like pop, R&B, or folk country – there are in any given song.
A user first selects a reference song that is similar to what they envision – for example, Don’t Stop Me Now by Queen. But they want it to be slower, with lower vocals, and fewer metaphors in the lyrics.
MyPart, the Israeli company behind the algorithm, will show the most relevant songs from the catalogs they have access to that match the chosen reference and music filters.
“People need music for specific commercial or creative needs,” says Matan Kollnescher, CEO and Co-founder of the startup. “And the way that they look for songs is still just like they did 40 years ago. They put on earphones, and listen to 1,000 mediocre, irrelevant pieces of music in order to get to one where they’d say, ‘wait a second, is this it?’
“Most talented people who work on these things know the right song when they hear it. The problem is that they are very likely not to ever hear it.
“That’s where we come in. We’re a song-search and matching engine, predominantly, based on deep analysis of lyrics, composition, and sound.”
MyPart is the only AI that is capable of categorizing music according to the content of the lyrics, as well as its style – whether it’s sophisticated or uses everyday language, and whether it rhymes or uses metaphors.
Its Songmine platform can also identify composition, or the chords that go into the song, and melody, the way in which the singer sings.
The user can adjust the percentages of each of the four categories MyPart analyzes – lyrical writing style, lyrical theme and mood, music and arrangement, and production and sound – according to what matters most to them, which will change their results.
“All of the other commercial engines that are trying to make music recommendations don’t analyze composition. They only analyze sonics, or how it sounds.
“And because of these capabilities, we’re really able to do things that up until today have only been done by really talented music executives, like find songs that might be a great fit for a commercial or an artist that’s looking for songs to sing.”
Sign up for our free weekly newsletterSubscribe
MyPart, founded in 2016, has started serving major record labels like Def Jam Recordings, Universal Music Group, and Warner Music Group.
“Music companies have different problems. They’ve signed a lot of artists and songwriters who are very talented,” Kollnescher tells NoCamels.
But their problem, he says, is that only a small chunk of them make it big. Labels like Universal Music, which is responsible for the success of artists like Taylor Swift, ABBA, and Elton John, have millions of songs in their inventory.
Songwriters who are part of these labels also have back catalogs of material, filled with hundreds of songs that have never been recorded, and many of which have been forgotten.
“How do you expect the music executives to remember every song in their inventory?”
MyPart is also going to offer its services to TV and film executives and ad agencies that could put in their own searchers instead of asking music companies for specific types of songs.
But it doesn’t serve people from these multi-million dollar industries alone. Small-time and unsigned artists and songwriters can use MyPart to pitch their work to famous artists, whose creative teams use the service themselves.
All the songwriter needs to do is upload the lyrics and an audio recording of their song. The lyrics, composition, and production are analyzed and ranked in accordance with what that famous artist is currently looking for.
When their team is searching for relevant songs, they have the option of using not just their label’s in-house catalog of music, but SongSource as well, the platform’s catalog for aspiring songwriters and artists.
“If they’re open to unsolicited, lesser-known artists, they will click SongSource, and assuming they are relevant to their search, they may consider these artists.”
Kollnescher had the idea for the company while he was studying at the Rimon School of Music, Israel’s largest school for the advanced study of pop, rock, jazz and R&B music.
“To me, it was pretty mind boggling how some of the most beautiful things that I’ve heard in life got nowhere. And I’m thinking about songs that were played in my class, or by people who have not been able to achieve commercial success.”
It saddened him that songs that touched him – some of which he doesn’t even remember – never got anywhere.
“So the original intention of building a machine that is deeply able to analyze and understand songs came from that, from the sheer volume of songs out there. If you don’t have something automatically doing this, things will get lost for sure.”