Jul 29, 2016
One man’s progress is another’s disconcerting future. While the gains are undeniable, the steady and inevitable march of technology can lead us to some uncomfortable places — artificial intelligence, the singularity, driverless cars or Pokemon Go players milling around our backyard.
As much as technology is linked to the history and development of classical music, it’s usually a product insulated from this kind of angst. You don’t see many science fiction writers expressing a collective cultural anxiety over a future that includes robot music directors or genetically engineered tuba players.
Or so we thought. A new study might point to a near future worthy of such apprehension. The study, published in early July by researchers in Germany, looked at how well the average listener could tell the difference between a performance by a real orchestra and a computer-generated one, and the results might be somewhat alarming for traditionalists.
How do you build a virtual orchestra? Sample libraries house collections of sounds recorded from real instruments that can then be arranged to simulate a performance of a piece by a soloist or full ensemble. These libraries have existed for many years, often used in modern composing programs to play back what has been written as a reference point for composers, but they’ve usually been primitive imitations of actual orchestral sound.
Technological refinement in the area, though, is inevitable. The high-end orchestral sound libraries today are a far cry from the MIDI-generated, nasally trumpet sound you might remember from a decade ago.
To look at how convincing modern orchestral sound libraries can be, German researchers wanted to see how well the average listener could distinguish between a computer-generated and real orchestra performance of Stravinsky’s Rite of Spring. The piece was chosen for its variety of instruments and techniques. In an online test, participants listened to 10 excerpts from both a real recording and one generated by Vienna Symphonic Library (one of the leading programs in this field) and asked to identify which was which.
Participants with various levels of music backgrounds correctly identified the source of the clip 72.5% of the time. That number dipped though when looking at the results of listeners without musical training.
“While sound experts tended to correctly identify the sound source, participants with lower listening expertise, who resembled the majority of music consumers, only [identified the correct source] 68.6% [of the time],” researchers write.
That second number might not seem significant at first. Someone without much musical background can identify a fake orchestra with a higher degree of success than guessing, which would seem to suggest a somewhat obvious difference. That may be, but in the philosophy of telling human from machine, the 70% benchmark is massive.
In 1950, British computer scientist Alan Turing outlined what he called an “imitation game,” which would later be called the Turing Test, to measure the intelligence of machines. The game involves human judges asking a series of questions in text-based conversation to both a human and computer responder for 5 minutes. The judge is then asked to decide whether they were communicating with another person or computer based on the responses.
Turing predicted that by the year 2000, a computer would be so adept at this kind of imitation that a judge would only identify their conversation partner correctly 70% of the time or less. Since then, 70% has become an important threshold in the philosophy of artificial intelligence.
So do the results of the German study mean that we have achieved orchestral artificial intelligence? Well, no. Don’t be silly. But they do suggest that computer-generated orchestral recordings are becoming increasingly sophisticated and perhaps soon, economically viable. For many in the orchestral world, that is a concerning thought.
Take the story a few years back about a Wagner festival in Connecticut that looked to use a recorded, virtual orchestra to accompany live singers in a performance of the Ring Cycle. The proposal created a firestorm. Singers received letters from pit orchestra musicians in Chicago warning them that “the live musicians of this country will remember you for the rest of your career and treat you as a traitor to our art form.” Participants backed out and the festival was eventually tabled. The specter of real life musicians being replaced by digital counterparts has become a heated topic, particularly for dance companies performing to recordings or on Broadway where electronic enhancement has been used to cut cost.
The idea that real musicians might one day lose an already-hard-to-find gig to a virtual violinist in a box is alarming. What the study shows is not that programs like the Vienna Symphonic Library are good enough right now to make a fool-proof facsimile but that they’ve come a long way. Who is to say how convincing they’ll be in another decade?
What should be a comfort is that classical music is a live performance business. What orchestras are selling is a chance to hear a performance of music in-person, in a visceral way that is an entirely different experience from hearing that same piece on headphones — even if that recording was made by the finest musicians in the world.
Just like with the artisanal coffee or craft beer movements, there is a market for quality, local flair and story. There are abundant, less expensive versions of those products available, but often those serve as entry points into the market that whet the consumer appetite for greater quality and experience.
Disruption is always worrying, but it also might offer some silver linings in the form of opportunity and access. For young composers who usually write for smaller ensembles because of the prohibitive cost of recording a piece with a professional orchestra, there is an opportunity to turn their sheet music into a semi-reasonable product that the layperson can hear. What would otherwise just be abstract notes on a page become listenable and concrete to friends, mom or even potential Kickstarter backers who might help that work get performed by live musicians.
For composers writing music specifically for computers, these advancements might offer an interesting opportunity to compose experimental works that might be physically and technically unperformable by human beings.
If this technology becomes viable, the biggest impact might be in the area of new recordings of film, TV or video game soundtracks. It’s hard to imagine a world where an established composer like John Williams would allow one of his scores to be performed by anything less than a live full orchestra. Big budget studios might always be willing to pay for quality, but it’s not inconceivable that a smaller production might cut corners by paying for live lead players supplemented by a virtual section of strings. At the same time, it’s possible that a composer for an indie video game, operating on a shoe-string budget that could never afford an orchestra, might now have the option to write something for the full orchestra sound.
The results of the study should give us all pause — not for what they might mean today, but for what they might mean in the future. Recordings are not as big a part of the business model as they once were, but they still mean jobs for some. Those concerned murmurs amplify to a riot when there is a possibility that canned music could outsource jobs in live performances.
But take that previously mentioned Wagner festival in Hartford, Connecticut. If the alternative to using a faux orchestra was to not perform the opera at all, is there a case to be made for a faux orchestra? Sure, it might be a lousy substitute, particularly for brass players like myself who listen to Wagner for those soaring horn and trombone lines. But what if the resulting ticket price means people who can’t afford a seat at The Met might hear The Ring? Would a few of those in that audience then decide they like it enough to go hear it in its full live-orchestral glory?
With a heated topic like this, there aren’t easy answers to these questions. It’s impossible to put the genie back in the bottle, and performers and artists must continue to grapple with and adapt to this evolving disruptive technology.