LISTEN: These AI-Created Sounds Have Never Been Heard Before
It may sound simple, but it's actually difficult to create a purely new sound. Well, chalk it up to another thing artificial intelligence can do that humans can't because an AI system is shown to be capable of inventing new sounds that have never been heard in the history of the world.
According to a report from WIRED, the team of Google Magenta -- a part of the company's AI laboratory Google Brain -- is using their AI systems to create entirely new noises using the mathematical characteristics of pre-existing sounds. It's not just layering one instrument's note on top of another's, but producing completely new noises.
The program is dubbed Nsynth (Neural Synthesizer), envisioned to develo new avenues of human expression, according to a blog post last month. Classic synthesizers use hand-designed component such as oscillators and wavetables, but Nsynth uses deep neural networks to generate audio.
The researchers first put together a massive database of sounds from thousands of instruments, which they fed into a neural network. The network created a mathematical vector out of the notes and characteristics of each instrument. Thus, the machine is able to mimic and combine the sounds of the wide variety of instruments available in its network.
"Learning directly from data, NSynth provides artists with intuitive control over timbre and dynamics and the ability to explore new sounds that would be difficult or impossible to produce with a hand-tuned synthesizer," the researchers wrote in the blog.
Critic Marc Weidenbaum told WIRED that the AI's approach isn't very different from the traditional practice of orchestral conductors.
"The blending of instruments is nothing new," he explained, adding that Google's technology could push this craft to new heights. "Artistically, it could yield some cool stuff, and because it's Google, people will follow their lead."
A paper on Nsynth is available online in arXiv.