• long running

Free: Nvidia RTX Voice (Background Noise Canceller) for RTX 20xx, GTX 9xx, 10xx & 16xx Series GPUs Compatible w/Workaround

2450

RTX Voice filters background noise from both audio input and output with Nvidia Tensor cores.

First, the good news: Nvidia's RTX Voice technology actually works really well. As you can hear in the test sample below, turning on the feature filters out almost all of the mechanical keyboard clicking picked up by a standard webcam microphone. That's likely to be an especially useful feature for anyone who has tried typing notes during a Zoom video call or dealt with kids screaming in other rooms during the same.

-Ars Technica Full Article

This noise-cancelling app is pretty incredible in action. I conducted a few tests last night before going to bed in which I ran random YouTube videos of conventions and other noise-filled areas through the filter. Sure enough, conversations were plucked out and put on center stage, with only the tiniest noise warbles indicating that background noise may have originally been there.
Reminder that the app works this way for your purposes, as well. Do you have a Zoom meeting with someone whose window is open or whose family members chatter in the background? Nvidia RTX Voice can be toggled to process the audio you receive, so you don't have to hear Jake in Accounting's 8-year-old muttering in the background ever again (even if Jake doesn't use this app).

-Sam Machkovech, another Ars Reporter

Officially, RTX Voice is only supported by Nvidia RTX 2xxx cards like the RTX 2060, but a workaround for GTX 9xx, 10xx and 16xx series Nvidia GPUs exists. Even older Nvidia GPUs with varying levels of success. Premodded .exe from Guru3D

Requirements:

  • Windows 10 (Also works on Windows 7)
  • Driver 410.18 or newer
  • Compatible Nvidia GPU

Supported Apps:

  • OBS Studio
  • XSplit Broadcaster
  • XSplit Gamecaster
  • Twitch Studio
  • Discord
  • Google Chrome
  • Battle.net Chat
  • WebEx*
  • Skype*
  • Zoom*
  • Slack*
  • Steam Chat**

Note:
* RTX Voice speaker output may exhibit issues in these apps.
**Turn off ‘Noise Cancellation’ for better quality.

Related Stores

NVIDIA
NVIDIA

Comments

  • +55

    For an example of how effective it can be: https://clips.twitch.tv/HeadstrongCleanDugongKappaPride

    • +19

      that's f insane

    • +3

      WTF, that's awesome.

    • It's mangled his voice to remove the background sound. Which you could already do using Reaplugs VST plugins for example. Presumably the background sound is sampled then that is subtracted from the recording following that.

      If you are dealing with normal levels of background sound such as fans it's better to use a gate followed by a downward expander. This basically removes all noise except for when voice is being recorded. The background sound is there still while voice is recorded but it gets drowned out.

      • +1

        This is an extreme example so it's probably not unexpected that the audio artefacts are worse as a result. FWIW I gave it a quick spin and noticed very little distortion when talking while clickity-clacking on my MX Blues.

      • +2

        I don't think I've seen a VST remove Keyboard clicks on the fly, as you can't provide them as a noise print like ambient sounds.
        Noise gates are great too but as you said, it's not going to do anything whilst you're talking, this does a very very good job of that.

      • +5

        A gate is a downward expander (with a high ratio - like a limiter is a compressor with a high ratio).

        It's not a great way to deal with background noise, because it can only 'gate' the channel shut when there's no signal. Once there's signal, it'll happily let through the background noise as well as your voice.

        The RTX noise removal is a different thing - it's actually removing the noise behind the voice.

      • +1

        REAPER FTW

      • This is also likely an almost perfect example of the uses of this software. I highly regular high frequency droning sound.

      • +5

        The biggest strength of RTX Voice is actually how good it is at dealing with irregular sounds. Example from the same guy: https://clips.twitch.tv/EnthusiasticPricklyGoblinDatSheffy

        There's a few more in-depth reviews floating around e.g. https://www.techpowerup.com/review/nvidia-rtx-voice-real-lif…

    • +3

      i'll have to clip the start of that it was super funny lol

    • +1

      Awesome Nvidia!!!

    • -3

      I heard somewhere that the guy in that twitch clip had to hunt for his food.

  • +8

    Workaround worked for me with a GTX 1060 6GB.
    Thanks :)

    • +1

      I only have the 3GB and it also works fine.

      And by fine I mean stupidly awesome.

  • Wow thats insane! i wonder if it works on post videos or just mic?

  • +3

    Linux support too please :)

    • +5

      So you are telling me you think Elgato is more popular than OBS? Got ya

    • The hack/makes it work on GTX cards, using the CUDA cores.

      RTX Voice works on all software, because it creates a new filtered audio input device, but Nvidia is not supporting all the software ATM in the same way that they aren't supporting the GTX cards that work with it.

      • +1

        Yeah It seems like the 'support' was just showing people how to switch the input and possibly disable excess configs in the app like its own post-processing. can't see why it would not work as an audio option anywhere in the OS.

    • ElGato doesn't produce streaming software.

      I have loaded this up in both OBS & Streamlabs OBS and it works brilliantly in both.

      Conclusion: You talkin' SH!T

    • +1

      A GPU is basically when it comes down to it a very large amount of low powered CPU cores. Hence the term GPGPU which stands for General Purpose Graphical Processing Unit. That’s often how they are used in HPC instances. It’s using them for different uses as per the usual for graphics.

    • GPU's can be used for more than render graphics. It's pretty exciting, google "GPU bitcoin mining" and prepare to have mind blown.

  • +1

    Gday, this is great. Works on desktop GTX1070 8GB and Lenovo E570 laptop with RTX1050 Ti 2GB. Upvoted.

  • +2

    Someone better post this "deal" too

    https://github.com/keylase/nvidia-patch

  • Does this require you to use your GPU audio or will onboard audio still work?

    • It works with any audio input and output.

  • +13

    I can confirm this works with output too not just input, i used it as a middleman between VLC and my headphones and filtered the TV SHow house, it removed all music and hospital sounds so all i heard was 98% clear voice/talking. works crazy well.

    • That's very cool for us muso nerds.

    • +1

      This is great for forcing your mates on discord/zoom to "use it" if they love a good noisy background.

      • I didn't think of this! Thank you

  • +2

    I have installed this with my GTX1070ti and find the results are pretty good on my Blue Yeti

    Samples:

    RTX Voice Off
    RTX Voice On

    • I couldn't hear any background in the off sample in the first place though

  • Any alternative for AMD users?

    • +1

      Yep. Download Krisp AI. Discord has incorporated into its program.

  • Finally, a WFM related use for my GPU.

    • +9

      Excellent. Now for the tax deduction on my 2080Ti.

  • +2

    This is a game changer for me, as someone who does professional presentations via video. I own a 5700XT and RX580, on my two systems, both will be replaced with Nvidia GPUs over the next few weeks, probably a 2070S or 2080S.

    Sure, this will cost me hundreds of dollars, but I was planning to further sound isolate the room which would have cost way more and perhaps not even been as effective.

    • -2

      Why not just use krisp.ai….

      • +2

        Unfortunately it's nowhere near as good. Also, there are other reasons I've been itching to switch over to Nvidia for too, e.g. the NVENC encoder. The quality of the AMD encoder is just not up to scratch.

        As great the value as the 5700XT is for gaming, it doesn't have the feature set of the Nvidia Turing cards.

        • -2

          Fair enough. I'll keep my RVII though :)

        • +4

          +1 on the NV encoder. I have an Nvidia GPU for Shadowplay and it works so much better than AMD's version of it.

          • +1

            @StrayfireX: I agree - I wish reviewers and press focused more on features than just FPS. My wife has a GTX 1080 and whilst my 5700 XT pushes more frames, she's happily streaming games to the Shield in our living room, recording using NVENC, using CUDA (which is more universal than OpenCL) for her statistical simulations and can now cancel me out of her voice calls. All of these features are even better on Turing than Pascal.

            I'm no AMD hater - I love my Ryzen CPUs and my RX 580 and 5700 XT have both been fantastic for gaming, but Nvidia have been killing it on the features and drivers recently.

            • @p1 ama: I have multiple shields and even the controller + Steam controller but haven't found a use case for streaming to my TV yet. I did it a few times and never again. What games does she play? I think I just like my 144hz 1440p monitor + keyboard/mouse, rather than comfort of the living room, ahaha.

    • +2

      You selling the 5700XT cheap for a fellow ozbargainer?

  • It's amazing how much noise it can remove. Just be mindful that the GPU will remain in higher power state while RTX voice is running, around 30w extra power consumption for a RTX 2070s according to techpowerup.

    https://www.techpowerup.com/review/nvidia-rtx-voice-real-lif…

    • and the ~10% perfomance hit when using mic input

      • Interesting. Wouldn't have thought it would be that high. Might have to compare in-game fps results

  • I've seen a lot of debating in the comments, but can someone explain what makes this different to traditional noise cancelling methods?

    Can someone also explain why this only works for 10 series and up at this point in time? Any reason a workaround isn't possible for Maxwell and below?

    • +2

      This is supposed to be "powered by AI" vs traditional methods, whatever that means.

      Maxwell does work but with "mixed results" according to the poster.

      I don't have a Maxwell GPU, can't confirm.

      I'm sure this could run on a CPU, but Nvidia wants this as a value-add for their GPUs much like CUDA acceleration or Nvidia Shadowplay.

      • Ah okay, that makes it more interesting. I'll have to check out those threads linked in the OP then. I've got a 970 and whilst I probably won't use this regularly, it'd be very interesting to try every now and then.

        • I'm 100% on your side tbh. I have a 1060 and it doesn't officially support this because Nvidia is putting bits of code into RTX Voice to stop it from working on 10xx and 16xx series GPUs. I only excluded the 900 and earlier because of reported issues.

          • @StrayfireX: I think it might just be that people who reported issues may have had something else going on or that they might not have done something correctly. I'm only guessing though, but I think it might be the case because someone on the forum thread with a 780 (Refresh of the 600 series 'Kepler' architecture) got it working. I haven't seen anyone with a 500 series or below try it, but I'm less inclined to believe the RTX cards have anything to do with it considering a Kepler reportedly works fine.

    • From playing with it briefly - traditional noise cancellation works best on background noise that is more or less constant (e.g. fans, electrical hiss) and they basically work by taking a profile of the background noise and subtracting it from the audio stream. Aside from not working well with intermittent noise, because the noise profile is being subtracted from the entire audio signal, as you 'subtract' more and more of the noise profile, you introduce distortion into the part of the signal you want (e.g., distorting someone's voice). My impression is that RTX voice is using deep learning to basically figure out what the actual noise signal is rather than assuming it's a constant profile, which allows it to subtract off even things that are highly transient (like typing). Traditional noise cancelling does a great job of getting rid of the constant noise but is not nearly as good at transient noise.

      As for 10 series…nVidia GPUs have different version of 'Compute capability' depending on the hardware - kind of like how CPUs have things like SSE or AVX. One early update in compute capability was to double precision math operations, for example. There is indeed a change in compute capability going from 9 series to 10 series https://developer.nvidia.com/cuda-gpus - whether it's about essential functionality or just performance though is anyone's guess, I'm sure there's a marketing aspect to it too though

      • Thanks for the explanation! I think it would make sense if it was using deep learning or something along those lines with how it is able to distinguish voice from noise, rather than a constant profile like you described.

        The GPU thing still confuses me. I would understand it if newer generations had different funcionalities to allow it, but then again judging from the op and linked forum thread success seems to be mixed. For example, someone with a 750 Ti reported it as working, so i'm not sure why other Maxwells may not work. I also saw someone with a 780 working, so 600 series/Kepler might also work too to an extent. Maybe the cases where it doesn't work had other things affecting it? Who knows… I wouldn't be surprised if it was just Nvidia trying to sell RTX, considering Kepler reportedly runs it fine.

  • Just gave this a try with my RTX2060, it is fantastic, absolutely amazing. It's very selective compared to just ordinary noise cancellation, presumably because of the machine learning. For example, it almost entirely suppressed typing noise for me without adding any distortion to speech, haven't seen anything perform this well with that type of noise before

  • +10

    ahaha, you guys should try it with big bang theory videos on youtube/netflix

    it cuts the laugh track out completely

  • Holy moly, works with my GTX1070TI! Played a music clip whilst talking, and I hear only my voice! It's craaaaaaaaaazy times.

  • GTX 1050, Zoom. Works with input (mic), but doesn't seem to work with output (speaker/headphones).

    • I have the same issue with my GTX 1060. Input but not output.

  • not working for me on win7 either.

  • Yes it should work very well for this, as a GPU is mostly just a very large amount of low powered cores. Think of it like the lowest powered intel atom cpu core you can think of then using parallelisation your generate large amounts of processing power.

    If you are interested look into GPGPU or General Purpose Graphical Processing Unit. That’s how they use them in HPC instances.

    • Not so much a large amount of cores, more like a decent amount of cores with huuuge amounts of general purpose registers inside each. As in 1024+ registers compared to only 16 in a regular x86_64 CPU. So the cores themselves are designed to handle much larger workloads, but suck at branching.

  • What about Microsoft Teams

    • Well if it works in skype there is a good chance it might work in skype, as teams is set to replace skype in the Microsoft family in the coming months / years.

    • +5

      been using it all day in MS Teams, was freaking my colleagues out by clapping and banging on pots and pans whilst talking to them. They all couldn't believe it haha!

  • +9

    No, I don't think this will work on wife.

    • +1

      I'm sure the deep learning that it runs upon could be trained to exclude particular types of interference, including wifely nagging, screaming kids, politician's lies ….

      It's giving me echos of Black Mirror's blocking of people and Hitch Hiker's Peril Sensitive sunglasses - use it as a filter for audio or visual input at all times and filter out those parts of the world you don't want to see or hear.

      Imagine an ad block that works for the real world at all times …

      • Oooh, thinking about it, it does give me vibes of that Black Mirror episode. You turn it on, it blocks you from hearing (those) people. Honestly it's kinda scary, how other aspects of Black Mirror which I thought were neat and way in the future are becoming closer and closer to reality.

  • -1

    Confirmed working on my laptop with an MX150.
    Definitely does not require RTX to function, despite what Nvidia claimed.

  • +4

    FYI this is great for conference calls etc but i wouldn't use it for gaming yet.

    https://wccftech.com/nvidia-rtx-voice-performance-impact-ben…

    Decent performance hit when using it.

  • +2

    Thanks for this, but I have an issue when I try to use noise suppression.

    "Unable to start microphone denoising"

    "Unable to start speaker denoising"

    I have a GTX 1070

    • +1

      I have stacked same, but solved it.
      Let me share how to solve,

      -dl and install lates ver of driver
      https://www.geforce.com/drivers
      -update driver using "geforce experience"
      -uninstall installed RTX voice (if installed once)
      -re-install RTX voice and failed
      -follow instruction on this page….

      Hope it helps..
      *Actually I'm the one who vote on this topic;)

  • Does this remove background voices too?

    • Only mine.

    • This is what I am curious about. If the kids come in to my office room and start yapping away, I'm assuming they won't be filtered out or will it work for this use case as well?

      • Nah it should clear up almost anything you can throw at it. Please note i am not a nvidia shill, just love using GPUs for processing or other stuff.

  • That’s insane. Anyone with one of those omni directional mics with a stand sitting behind a loud mechanical keyboard give their feedback? I’ve always wanted one, not for streaming but voice clarity. Having it sit infront of the keyboard was impractical so never thought more of it. This is a game changer

  • This is fantastic.
    Perfect for people with noisy keyboards.

    But it filters other people's mics as well, haven't tried it yet but I have a few people on discord I'd love to test this on

  • I'm not that savvy with the different Nvidia cards and costs. Having a google I find the cost of an RTX2060 (~$2k) is waaaaaaaaay out of my budget.

    What's the cheapest path (ie card) to get this feature? I dabble in video editing as a hobby and will probably upgrade my graphics card soon.

    Edit: Bad google / idiot user - I rechecked…RTX2060 is about $600 which is not that bad. I must have stumbled on another model.

    • GTX 1660 Super will work with workaround.

    • If the workaround works, which a lot of evidence supports is does, it should work on as low as a GT 1030 for ~$150, a comment above states "Confirmed working on my laptop with an MX150" which is a virtually identical laptop variation of the GT1030. Keep in mind second hand cards too, any Geforce 10 series should work.

    • +1

      A GTX 960 works, i got that for under 100

  • Jake in Accounting at Ars is probably shaking his fist at Sam right now…

  • My thinking of how this works is not that it subtracts the noise, but that it completely synthesizes the voice. It extracts something that characterises the voice, and then recreates it from that characterisation.

  • Works for my GTX 960 4GB

  • +2

    Just put together a vid on my YT Channel.
    Did some intense demo's with banging a pot, steaming a Shirt and running the dryer in the Laundry haha, but this stuff is amazing for general keyboard clicks and heater/air-con noise.

    total game changer

    p.s. also show my routing for VM Banana if you're a External Audio interface user wanting to update your routings.

    • +1

      I like your channel mate! Subbed :)

Login or Join to leave a comment