Tuesday, May 31, 2016

SoundCloud’s free “auto-mastering” audio tool is more of an auto-turd

Written by Sam Machkovech — Grammy-winning engineer: "If anyone thinks it's a good idea, they get what they deserve."

First, the robots came for our factory jobs, then for our fast-food jobs. Now, they're aiming their laser-guided sights at... music-recording engineers?

On Thursday, the audio-processing company Landr (founded in 2014) announced its partnership with the hugely popular self-publishing music platform SoundCloud. SoundCloud users can now have their original tunes processed and "optimized" for free by Landr.

This news promises a different kind of audio-related snake oil than we're used to at Ars. We've seen plenty of products advertising instant audio benefits, from cords to pre-amps to DACs, which largely target music consumers. Landr flips its sales pitch by targeting music creators.

Landr's landing site describes the mastering process as "complicated and elusive," then insists that its product, which is almost entirely algorithm-driven, delivers a quality product for small-fry musicians by intentionally limiting how many options they can pick from. "Great design is all about limiting the field," Landr says. As a result, the company touts that "we're confident you'll hear the difference" between professional mastering work and what Landr can pull off.

After our tests of SoundCloud's new Landr functionality, we can safely agree with that statement—in every bad way possible.

Landr takes on spacey drone and cyber-thriller techno

Before taking the service's new auto-mastering wares for a spin, we reached out to a few music creators, producers, and engineers for their thoughts. The biggest-ticket response we got was from Grammy-winning engineer John Congleton, who's worked with St. Vincent, David Byrne, Wye Oak, Okkervil River, and many more. Congleton laughed Landr off in no uncertain terms: "I'm not getting up in that shit. If anyone thinks it's a good idea, they get what they deserve."

The first music creator to respond to our public request for uncompressed audio was experimental "drone" musician Steven Flato of San Diego. At first, his combination of classical strings and woodwinds with intentionally deformed electronic effects didn't seem like a good fit. That's a lot of sonic soup to put under the Landr microscope, which is when we realized he was an ideal choice to test Landr's bold claims about technologies like "micro-genre detection."

Flato sent us two versions of the opening track from his EP Exhaust Sytem: the final version, mastered by his longtime recording partner A.F. Jones, and the unmastered track that Jones had to work with. All of Flato's files were uncompressed FLAC. We dumped the unmastered version into SoundCloud and connected that site to a new Landr account. After we watched the site toss up a bunch of distracting, jargon-filled loading screens, it served up an unlisted SoundCloud link with Landr's auto-mastered take on Flato's composition. We were also given a single free WAV download as a new user; additional future uncompressed downloads would cost a minimum of $5.

Landr offered us "low," "medium," and "high" mastering presets without any explanation as to what was lower or higher about those options. We opted for "medium," since that's the site default. The interface includes an A/B test of a given song's final 30 seconds, split between the original file you upload and its mastered version. Both come as 192 Kbps MP3s. Flato's track has a very quiet ending, so this specific sample didn't do us much good. We instead queued up all three tracks—the two we received from Flato and the uncompressed result from Landr—on VLC, which supports very easy jumps to exact points in a song for listening comparisons.

Normally, this sort of testing is very dangerous territory. We have yet to buy a complicated rig of microscopic microphones to arrange in our ears, like some sort of piercing array you'd find on an extra in the film Tank Girl. That leaves us with only one boring option: giving you, dear reader, our biased, subjective take on how a thing sounds to us. However, Landr's version of Flato's composition saved us some trouble.

Whatever Landr has cooking in its "micro-genre detection" algorithm failed to acknowledge that Flato's music lacks anything in the way of vocals, drums, rhythm, or even key changes. Landr forcibly extracted as many sounds from the intense mix as it could, all while smoothing and compressing the track to an incredible degree. Certain intentional artifacts that had been raised in the sound profile by the professional master, including a noticeable, tension-building rattle, were smoothed out by the auto-master. Landr's version also drew out and isolated some buzzing, sinusoidal waves of noise that fit better in the professional master as a supplemental tide. The result nerfed all of the original track's intense, experimental qualities. Landr's version felt like the equivalent of someone going into a Li'l Jon track and grimacing at so much bass. It missed the point.

Jones, who mastered Flato's EP, was not impressed by Landr's take. "It is pretty scary," Jones wrote in an e-mail interview with Ars. "There are hyper-compressed areas that have resulted in clippy digital artifacts. In other areas there are certain noises that are actually enhanced when they should have been reduced or eliminated. The equalization is wrong. I could go on."

The only other track we received before going to press was from an old friend in New Jersey who records under the moniker Kursse. His dark, synth beats sound like they've been ripped from a cyber-thriller film or video game (and as a result, I imagine more than few Ars readers would be into his stuff). Kursse was kind enough to provide both the unmastered and pro-mastered versions of his song "Demonizer." Considering the song's serious bombast, we opted for the "high" auto-master setting when we ran the unmastered version through Landr.

The Landr version makes some weird sound-alteration choices. Most notably, it strongly favors the high-end in its melody while flattening everything else behind it. The biggest casualty in the Landr take is the aural space that the bass-drum sounds enjoy on the professional master, in which the mix leaves room for those sounds to swell and recede for maximum loud-system and headphone pop. Landr instead takes the song's dials and cranks them all to 11, which kills the impact of every new techno-sound movement introduced during the song's 3:34 runtime.

Would you trust Siri to copyedit an article?

"One of the first topics of discussion a client might have with a mastering engineer might involve talking about influences, or perhaps reference material such as a particular song or album with production quality that the band really appreciates," Jones said to Ars when asked about his mastering process. "This can often be crucial information and a good launching point for the multitude of decision points that are available to the engineer in working with their client's music. I can think of no such option available that could possibly guide an automated process."

Jones' brief tinkering with SoundCloud and Landr left him wondering why the service was so insistent on its less-is-more philosophy: "If there are certain things I like about a Landr master—and thus would like to retain—but other areas that are in need of specific improvements or complete changes, it is important that the musician is able to communicate those exact things and then to expect specific results based on those parameters."

According to Jones, DIY musicians who lack cash and expertise but still want properly enhanced sound are better off using any number of currently-available open-source software to do their mixes. Interestingly, Landr's free and comprehensive guide to the wide world of digital audio workstations (DAWs) encourages musicians to explore settings instead of relying on automatic or extreme results. "Be careful not to get carried away," Landr's guide warns. "Applying too much compression is a danger zone. Using only compression to try and balance levels in a mix will lead to a lifeless, punch­less, and fatiguing mix. Yucky."

Jones had a lot to say about why automation, especially Landr's take on it, is dangerous for a song's final phase. But his closing statement about his expectations for future sound-processing automation was perhaps the most telling: "For mastering or any other trust based phase an artistic process there is no substitute for ears ideas in human interaction." That was his copy-and-paste of how Siri processed his spoken sentence, which lacked punctuation and misidentified or skipped words. His e-mail interview concluded with this question: "Would you trust Siri to determine the final text for this article?"

I do not share Jones' fatalism about what automation can bring to industries based on optimization. Supercomputing, combined with enormous databases, may one day lead us to a music processing system that is genuinely aware of genre and context. But I still agree that any such system will need more user interaction to make sense for a creative enterprise. As for Landr—this two-year-old review of the service's launch reads like it was written yesterday. It concludes that "Landr will get better over time." How much time remains to be seen; for now, SoundCloud and Landr users can expect to get what they pay for.

Click here to read from this article's source.