Advertisement

Audacity now has a free AI-powered noise suppressor but the machine isn't going to replace a sound engineer anytime soon

 Focusrite Scarlett 2i2 Studio bundle microphone on a Rode mic stand.
Focusrite Scarlett 2i2 Studio bundle microphone on a Rode mic stand.

Audacity has become the absolute go-to recommendation when it comes to free audio recording software for your PC, and while it's an easy program to start recording with, the built in noise reduction in particular has always been a bit of a bugbear of mine. It's often prone to introducing a lot of digital artifacts into the signal in an effort to get rid of background noise, so when I heard that there was now a new free suite of AI tools based on OpenVINO, including an AI noise suppressor to play with, I had to give them a try.

OpenVINO is an open-source toolkit that's built for running machine learning on Intel desktop and cloud platforms, and these tools have been built by Intel within the platform for Audacity (via CDM).

Beyond the AI noise suppression tool for podcasts, there's also a "Music Generation" tool designed to generate music from a prompt, or based on existing music, and also a "Music Style Remix" feature that can split a song into its component parts.

It's the noise suppressor in particular that caught my interest as I've had some previous experience engineering a podcast or two, and I always sought to use expensive VST plugins for background noise removal, as the Audacity default effect can be a bit of a blunt object. It's not awful by any means, but it often introduces digital "wash" into your vocal recordings that can be tiring to listen to.

With that in mind, I downloaded the new tools from Github and got to testing. Grabbing the nearest cardioid condenser mic to hand, I first recorded a test clip in Audacity with a noisy background, in this case my PC fans and a large, open and echo-prone room with a window facing out onto the street.

I then recorded another clip, and post-processed it with Audacity's built in noise reduction used at a high enough setting that the background noise was removed, before recording a final test clip and trying out the AI tool to see how it compared, and you can hear my results below.

Hmm. While the AI tool can absolutely remove background noise well, I'm not a particularly big fan of what its done to the overall tone of my voice. Listening to the Audacity noise-reduced clip side by side, while you can definitely hear more of that undesirable robotic-sounding wash it has at least kept some clarity in my dulcit tones, whereas the AI tool has made things a little muddy.

There's very little tweaking available for the OpenVINO noise suppressor as things stand, as it only currently has one noise suppression model included and no factory presets or sliders, so for the purposes of this test I had to run with the default settings provided. There is the option to import user presets, so I imagine it's a watch this space to see what the community might do with it.

A screenshot of OpenVINO noise reduction settings, or rather, the current lack of them
A screenshot of OpenVINO noise reduction settings, or rather, the current lack of them

Stream machine

Best microphone for gaming: make sure you're heard
Best webcams: be seen while you get your stream on
Best capture cards: lessen the load with a dedicated card

Still, the noise reduction itself is impressive at least, and it processed relatively quickly on my CPU, although it has to be said that the built-in Audacity tool is still quicker, by which I mean, near instantaneous on such a short clip.

You could absolutely EQ the AI clip to gain some clarity and tame some of the muddy notes in those mids, but then that does bring into question why you'd go to the extra effort unless you were really desperate to remove every last trace of background noise possible.

Needs work, I would say. I did attempt to play around with the AI music generation tool to get it to produce something presumably fantastic around the prompt of "PC Gamer Hardware team funk", but alas, no matter how I tweaked I couldn't get it to generate anything but an error message, even when reducing my prompt down to plain old funk, rock or pop. Shame.

Still, these AI tools are free to play with and seemingly still under development, so I wouldn't give it long before they're likely to be much more workable. While we now live in a world where the term "AI" is seemingly bolted on to every new product, something like AI-powered noise removal strikes as a good use of the tech, and given that its for free I'd say it wasn't a bad result.

Still annoyed about the PC Gamer funk failure though. One day we'll get our theme tune. One day soon.