How Landr is bringing machine learning to the music business

While it can’t yet replace a professional recording engineer, this Montreal startup is making mastering cheap and easy for amateurs


Woman using music mixing software at a keyboard with headphones on

(Peter Muller/Getty)

Before a hit song ever reaches your ears, chances are it’s gone through a process called mastering. It’s a somewhat mysterious and rarefied technique for outsiders, one performed by trained audiophiles who essentially fine-tune a mix so that the finished track sounds loud, clear, and consistent with other songs on an album. Mastering is as much art as it is science, but Landr, a Montreal startup, believes its software can do the job just as well as any human. “I don’t think I’m going to ever argue we’re better than the best mastering engineer in the world,” says co-founder and chief creative officer Justin Evans, “but I do know that we are perceptively close.”

Landr deploys big data analytics and machine learning to analyze a track and determine how to adjust parameters such as compression and equalization to make the song sound as pleasing and dynamic as possible to the ear. Musicians can sign up for a subscription with Landr, drag-and-drop their song files into the service, and receive what the company calls a “professional” sounding mastered version within minutes. The technology was developed in part by Stuart Mansbridge, a student at the Queen Mary University of London, who had been researching machine learning applications in the music industry. Evans, who previously ran a design and marketing agency in Montreal, was looking for a new business venture, and connected with Mansbridge a few years ago. As a former musician with an interest in machine learning, Evans was fascinated by Mansbridge’s work, and the two decided to commercialize the technology. (Pascal Pilon, a Montreal tech executive, later invested in Landr and joined as CEO.)

The company released the first version of its product in May 2014, and has since mastered more than 1.5 million tracks. The primary customers are amateur musicians for whom a professional mastering job—which can cost thousands—simply isn’t feasible. “The ability to make this possible for everybody is radically transformative,” Evans says. Shortly after launching, however, Landr realized that music labels and recording studios could be customers, too. TRI Studios, founded by Grateful Dead guitarist Bob Weir, has used Landr, and Warner Music Group led a $6.2 million investment in the company last year. (Landr now employs 56 people in Montreal and Los Angeles.) One reason for the interest, Evans says, is that labels are sitting on large back catalogues and live recordings that might not be economical to professionally master, but become commercially viable when processed through Landr. And although professional musicians are not the target market, Landr could still prove useful, particularly for hip hop artists releasing mixtapes. “Hip hop is so competitive, so if you can get a really awesome sounding master for a tenth of the price and put [the savings] into marketing, maybe it makes sense to use Landr,” Evans says. (He adds that a number one record has been mastered with Landr, but says he’s not at liberty to name it.)

As you might expect, Landr has proven to be controversial with professional mastering engineers. Some are intrigued by the concept (such as Frank Vasquez, a recording engineer with Snoop Dogg’s Doggy Style Records) while others view the claim that software can approximate human mastering as downright offensive. Steve Albini, a prolific recording engineer in Chicago who’s worked with Nirvana, Pixies, and the Stooges, called the concept “preposterous” in an email. A mastering engineer, he explained, is supposed to translate the wishes of the musician into a finished product. In short, the end result depends entirely on what the client wants. “There is literally no way it can be automated any more than you could automate getting your hair cut,” he wrote. Noah Mintz, the owner of Lacquer Channel Mastering in Toronto, wrote in Professional Sound magazine last year that Landr “misses the point” of mastering. “The human element is what makes mastering, mastering. It can’t be reduced to an algorithm.”

Like many mastering engineers, João Carvalho, who operates a studio in Toronto, has been curious about Landr. He recently compared a track he mastered to one processed by the company’s algorithm. Perhaps unsurprisingly, he wasn’t all that impressed. “It doesn’t sound that good, that’s for certain,” he says. Carvalho found Landr lowered the overall volume of the song while strangely boosting the bottom end to an unpleasant, speaker-quaking level. “This tool does have its place,” he says, “but I thought Landr would be a little more advanced.” A reviewer for industry publication Ask.Audio was more charitable, writing “this latest incarnation of Landr surprised me with its quality.”

Evans says Landr’s machine-learning technology is constantly improving; the more tracks it processes, the more nuance it can apply. He also understands why some mastering engineers take umbrage with the concept. “It’s like asking taxi drivers how they feel about Uber,” he says. The company isn’t out to replace human engineers, however, and Evans says there will always be a role for professionals. He likens the concept to autofocus in cameras—a high-tech development that enabled amateurs to take better photos, but also pushed pros to experiment with the art form in new ways. “We’re the people who terrify everybody right now, but we’re pretty friendly and care about the quality of sound,” Evans says. “We’re good people to be doing this.”