RX 9 izotope

Artificial Intelligence in Music Production – Friend or Foe?

During the 2010s we witnessed the rise of smart and fast machine learning algorithms, and their gradual integration into music production software. Now the artificial brains of countless plugins, platforms, virtual mastering suites – and even composers – are readily at the wheel as the cognisant drivers of today’s creative tools. But, has the balance between AI and human skill been tipped too far?

“Once the machine thinking method had started, it would not take long to outstrip our feeble powers. They would be able to converse with each other to sharpen their wits. At some stage, therefore, we should have to expect the machines to take control.” So was the bleak forecast that the father of artificial intelligence, computer genius Alan Turing predicted. While, thankfully, we’re not quite facing such a nightmarish end-point just yet, Turing’s notion of a machine-driven cyber-brain, continually sharpening its senses via constant dialogue and refinement, is a concept that lay at the heart of the commercial application of artificial intelligence.

In our industry, we can see this perhaps clearer than any other. Multitudes of AI-based software tools now available, that excel in tasks (such as frequency editing, mix separating, audio restoration and mastering) that it would take even highly skilled human operators a much, much longer time to undertake. Constant refinement, and adaptive learning routines, allow AI-based software pathways to improvement. While the term ‘artificial intelligence’ is typically ascribed to any software that relies on algorithms to fulfil its criteria, there’s actually quite a wide spectrum of definitions. There are those that simply trigger a series of pre-determined actions that its creators have carefully crafted, and then there are those HAL-like virtual geniuses (iZotope RX 9, Zynaptiq Adaptiverb for example) that can inspect a waveform, precisely diagnose what needs to be done to bring the most clarity to it, and take immediate action.


While these types of applications are perhaps more palatable, the increase in virtual composers of music has unsettled some. Initial experimentation with human-free music creation began as far back as the late 1950s, when University professors Lejaren Hiller and Leonard Issacson used an early computer to program the Illiac Suite (String Quartet No.4). This first foray into algorithmically-driven music creation nudged open a door that ensuing pioneers widened, leading eventually to 1997’s ‘Experiments in Musical Intelligence’ program. This was able to exceed human composers by generating a piece of music that matchlessly replicated the style of Bach.

While these toes in the water built the foundation for research, the last twenty years has witnessed an exponential explosion of compositional AI development, in conjunction with the rising speed of computer power. Now the options are vast. There’s Aiva – a classical and symphonic virtual music composer, which uses neural networks to scan huge libraries of classical music and replicate the commonalities it encounters. There’s Amper, which is able to conjure an infinite number of ready-to-go soundtracks for video games and TV, then there’s Loudly AI Studio, able to generate a range of tracks based on modern genres instantaneously.


While some of the results can be utterly superb, is the growing range of AI-based composers leading us step-by-step into a world where we’re surrounded by facsimile of human produced art? Aiva’s CEO, Pierre Barreau told The Naked Scientists that “Even if AI is objectively get better at composing music than humans, I think one crucial element that humans bring to the table is meaning in what they do. And an AI could come up with a new style of music, totally crazy style of music, but if there’s no creative intention that can be explained, I think it’s very hard for an audience to really connect.”

AI composition might be an easy – and substance free – solution for those needing the artifice of a pro-sounding soundtrack, without forking out the cash. But it’s understandable why so many jobbing professionals feel like their stand in the market is being devalued. It’s a debate that will undoubtedly continue.


Separate from their stance on the human vs machine debate within the compositional domain, many music producers have happily integrated AI and machine learning plugins into their workflows, aimed at speeding up and handling previously time consuming processes. Audio repair is one such field in which AI has delivered majorly impressive results, with world-leading companies such as iZotope proudly putting their reliance on machine learning as a front and centre USP of their product line which includes the audio post production package RX 9.

Melissa Misicka, Director of Brand Marketing at iZotope, explains to us that using artificial intelligence to this end was always a company ambition; “One of our goals as a company is to find ways of eliminating more time-consuming audio production tasks for our users so they can instead focus on their creative vision. Introducing assistive tech —that can intelligently analyse your audio and provide recommended starting points — felt like a perfect way to do that.”

RX 9 izotope

It’s not just about making time-consuming processes quicker, though. Many see AI as a method of achieving those tasks that humans are largely incapable of doing. iZotope explain how this idea has been put to use. “One example is source separation for speech cleanup.” Misicka tells us, “Our Modules like Dialogue Isolate or De-rustle rely on it to attenuate unwanted sounds like footsteps, bird chirping, or rustle of a mic hidden in clothes. Manual repair of these noises would be very laborious, because the noises change in time and overlap with speech”

“Another example is smart synthesis of replacement sounds.” Melissa continues, “When speech is coming from a telephone call, its frequency spectrum gets limited to 4 kHz, which results in a characteristic muffled sound. RX’s Spectral Recovery module uses machine learning to recreate the missing upper frequency band with realistic synthesised content to enhance the quality of speech. Manual ways for high-frequency synthesis would include tools like an exciter, but the quality and plausibility of the synthesised content would be nowhere near the results of machine learning.”While the arguments over artificial intelligence in the compositional domain continue to rage, few would object to the harnessing of machine learning to fulfil tasks that are largely outside of most of our aural and technical capabilities. Would they?


One of the most beneficial applications of AI for home musicians has been the quick availability of algorithm-driven mastering services. Take LANDR for example, this subscription service’s shrewd software leans on a mine of intelligence cribbed from 20 million mastered tracks. It uses this information to calculate how it applies tailored frequency-boosting and aural gloss for your song. “When LANDR first launched in 2014, it was a first of it’s kind solution for cloud-based AI mastering”. Patrick Bourget, LANDR’s Product Director, tells us. “In 2016 the landscape began to see similar but far less refined alternatives emerging in the marketplace.”

LANDR is an Artificial Intelligence driven platform

Bourget observes that many companies point to AI without justification, in contrast to LANDR’s always-evolving processes; “We see many companies leaning on the ‘AI’ buzz word, but they rarely deliver on their promise of truly intelligent production tools. LANDR maintains that our AI remains at the cutting edge of the AI field with one-of-a-kind results, every time—our engine adapts to a track’s unique sonic qualities when mastering. We never use presets to make cookie-cutter masters.”

“We’ve always felt that providing too many presets or options to users diminishes the value and trust we’ve built over the years with creators.” Patrick continues, “Our superior-quality masters stem from AI informed by millions of mastered tracks and tuning provided by the golden ears of industry giants.”


With the everyday prevalence of platforms like LANDR, Patrick Bourget seems like a good person to ask about how he sees this human/machine dynamic evolving in the future, specifically in the mastering domain; “Given the accelerating pace of creation and the often tight budgets of music producers around the globe, we feel that there will always be room for both AI-mastering & mastering engineers.” Bourget explains, “We’ve heard from countless professionals that LANDR provides an affordable and elegant solution for quality masters at a fraction of the cost of traditional mastering.”

LANDRs Artificial Intelligence-driven mastering styles

But what about those that fear their livelihoods could come under threat by AI’s continual advances? Bourget takes the middle ground; “We don’t see AI-mastering as a question of OR, but rather as an AND that assists creators when needed. Our automatic mastering process and simple workflow gives musicians the power to complete a master in minutes. An example being that of a mixing engineer quickly delivering a mix in progress with the polish an artist expects to hear from a mastered track. It’s a great way to get quick feedback.”

iZotope echoes this fundamental point, that the most successful applications of artificial intelligence to date are those that help creatives and professionals meet their objectives, and not those that seek to supplant them. “We often imagine our assistive tech as, quite literally, a studio assistant who can take that first pass at repairs or a mix for you while you go get a coffee.” Melissa explains, “We’d reinforce that the mission of iZotope’s assistive tools is not to replace professional expertise, but to coach those who are still learning by suggesting next steps, and to assist those who are more experienced by getting them to a starting point more quickly.”

While it’s unarguable that artificial intelligence will continue to soak into our daily lives on many levels, it’s plainly apparent that rather than shrinking before its fathomless potential, musicians and producers have more to gain than to lose from its ever-developing abilities.