Landr wants to conquer music mastering using machine learning
The ability to make music has always been a distinctly human trait. Yes, whales whistle and wolves howl, elephants use low-frequency rumbles to communicate and woodpeckers drum but none of them sound like Tina Turner or Leonard Cohen, able to command thousands of their species into packed stadiums or compel thousands of dollars to be charged up on iTunes.
See also
You would imagine that machines are similarly incapable of capturing the minds, hearts, and wallets of us humans. After all, it is one thing for AI to replace human occupations that are tedious, repetitive, and not particularly high on the value chain of human ingenuity. It is an entirely different thing to have a machine come up with something that can delight us humans with innovative brilliance, quirk, originality, and feel -- or even shape the sound of a song that can transport us to an emotionally complex place.
Landr, a Montreal-based company, wants to upend that notion by automating the intricate and complicated task of mastering audio tracks.
Normally, mastering occurs in the final stages of the production of a song when it is tweaked and adjusted so that it sounds not just clear and consistent, but also richer, fuller, and true to what the artist had in mind and sans the inevitable audio blemishes that crop up during recording. Like producing hand stitched leather bags from Italy or a complex Burgundy wine, mastering has always been thought of as a process that is part art, and part science.
Now, thanks to an algorithm that benchmarks and analyzes a large trove of songs that have been previously mastered, as well as songs in other genres for similar patterns, customers can simply plop their raw songs onto Landr's cloud engine which then rips through the process and delivers it back to the customer in a jiffy. Online music hosting site SoundCloud has been impressed enough to forge a partnership with Landr for its customers.
There has been a predictable uproar from music aficionados and sound engineers, lambasting the service for using the same algorithm to master all genres of music, for its tinny sound and its inability to navigate nuances -- its fundamental lack of 'musical understanding' as described by one industry hand. The technology site Ars Technica provides this scathing review.
Yet, this hasn't deterred co-founder and chief creative officer Justin Evans, who says, in Canadian Business, that he feels their pain but ultimately "it's like asking taxi drivers how they feel about Uber."
Evans suggests that the way to look at Landr is in the same way we first began to view autofocus in cameras, which allowed amateurs to improve their pictures but which also prodded pros to try new things.
I think a more accurate analogy would be the smartphone itself (and its camera), now so ubiquitous and sophisticated that taking photos is no longer the purveyor of the privileged.
It is in this respect that I think Landr makes a case for itself. Apparently, only about one percent of music created ever gets mastered, because it is such an elaborate and relatively expensive process. Now, it sits within a mouse click's reach of all of struggling musicians with stars in their eyes and songs in their heads.
Being able to master your song -- however inexpertly it may be done -- on the same day that you recorded it is a sensational leap, and music to the ears of professionals and amateurs alike. Its promise reminds me of the revolution that 3D printing has brought to the world of manufacturing by drastically cutting short the design-to-testing loop.
In this case, Landr allows you to record, master, listen, tweak, experiment, and redo, which could allow for all sorts of sound innovation in time, when the service has improved its algorithm and is able to detect genres and more complex mixes with better results.
It's bound to happen. Just a few years ago, a jazz-bot birthed by Sony's Computer Science Lab in Paris came up with a tune that was an amalgam of Charlie "Bird" Parker and French composer Pierre Boulez that stunned the jazz community with its sophistication and verve.
Most recently, Sony's AI system (called FlowMachines) also came up with a pop tune, 'Daddy's Car' that was widely regarded to be a pretty decent imitation of a Beatles number, accomplished by first analyzing a large cache of songs and then drilling down to a particular musical style. The Fab Four it ain't (incidentally, French composer Benoît Carré arranged the songs and penned the lyrics), but even the most defensive of luddites will admit that the song isn't half bad.
Which makes Landr's innovation the logical next step in the evolution of music in a machine-saturated world. Moreover, despite all the outcry, some of the big dogs in music are flocking to the Canadian company, with artists such as Tiga, Richie Hawtin, and Nas having employed its service. I'm not sure what Jerry would make of Landr, but Grateful Dead guitarist Bob Weir's company TRI Studios has used it. Music industry biggies, including Warner Music Group, have collectively pumped $6.2 million into the company.
There's another reason why Landr is music to the ears of successful artists and music studios. As Canadian Business magazine explains, both tend to possess vast quantities of back catalogues and live recordings that don't make business sense to throw professional mastering money at, but which can now become viable products when pushed through Landr.
It also points to the ultra-competitive field of hip-hop, where struggling artists can now re-deploy much needed cash that was once earmarked for mastering into marketing, which is where the real battle for success tends to lie.
So, for now, the intersection of machine learning and music appears to be somewhat beneficial, if largely imperfect. It is when we start rocking out to a Top 40 tune produced by the 'Spice Bots' that we may need to radically re-evaluate our position in this world.
But that situation may be just around the corner.