Okay I've just come across the new software by Nvidia and been playing with it on my live stream and it is incredible.

Then I thought it must improve speech recognition, and without doing much testing it seems to have made a huge difference!

Check out this video: https://youtu.be/Q-mETIjcIV0?t=536

I started it at the most extreme part.

Here is a link to the software and it helps if you do have an RTX card, but you don't need one.

It works with any microphone, it also gets rid of background noise in terrible YouTube videos.

Linked to the software: https://www.nvidia.com/en-us/geforce/guides/nvidia-rtx-voice-setup-guide/

Wow, it sure cuts out background white noise! And the video does a good job of explaining how to use it on non-RTX-card systems. Be sure to read the video description for a correction note.

-------------------------

Win10/11/x64, AMD Ryzen 7 3700X/3950X, 64/128GB RAM, Dragon 15.3, SP 7 Standard, SpeechStart, Office 365, KB 2017, Dragon Capture, Samson Meteor USB Desk Mic, Amazon YUWAKAYI headset, Klim and JUKSTG earbuds with microphones, excellent Sareville Wireless Mono Headset, 3 BenQ 2560x1440 monitors, Microsoft Sculpt Keyboard and Logitech G502 awesome gaming mouse.

The more I have been using this more amazing it is.

I can see this being built into the next version of Dragon NaturallySpeaking.

They would probably have to pay royalties to Nvidia. Slim chance of that happening.
Zig

I don't have an RTX card, but tried to install it with the workaround described in the YouTube video. Didn't work unfortunately, however I'm getting a new video card in the coming weeks so will try again. My help to cut out the noise of the kids fighting in the background when I'm trying to dictate.

Quite amazing! I have a Quadro T1000, do you think the software is compatible out of the box?

I wonder if it will help with the MS Word jumping cursors problem

-------------------------

Win 10 - DPI 15 - AT 8 pro + Andrea USB

this sounds interesting, i am in much need to have something remove background noise when using Dragon, but i am afraid to try it with this much information only. Woudl be happy if someone would write down instructions... and how do i know whether it will work on my computer or not?

With the suggested workaround, I was able to install the software, but I couldn't check the box at 'remove background noise', because that gave an error message 'unable to start microphone denoising', or something like that. So I assume there's no benefit for me to using this.

I have a recorded audio file with a lot of background noise.  I was wondering if there was some way I could use this software to clear that up using a virtual cable or something else by playing the audio file again and recording it with this NVidia RTX Voice filter on.  Is this possible and if yes how do I go about this.

We have experienced a lot of with this idea and driver in the past. Although you may eventually make the "trick" work with a few cards other than those from Nvidia, it will probably not work in most cases, neither be too legal to do so.

 

This is why we are currently exploring the possibility of integrating such a functionality in our Octopus USB Controller application and/or our TableMike and TravelMike microphones.

 

Let us know whether you may be eventually interested in such a development through this thread for us to allocate more resources and priority to it, since we are only focused in bringing to the Speech recognition market UNIQUE and value added solutions!

-------------------------

SpeechWare Team

 

 

 

Windows 7, Intel  i7, 8 GB, USB TravelMike or SpeechWare USB MultiAdapter with FlexyMike Dual Ear

Yes, I sure would be interested in such a development. Thanks.

Thanks for your feedback.

 

The special function that you are proposing for this denoising audio functionality irrespective of your audio card, i.e. filtering an existing recorded audio file, is also quite interesting for our project and we will take into consideration...

-------------------------

SpeechWare Team

 

 

 

Windows 7, Intel  i7, 8 GB, USB TravelMike or SpeechWare USB MultiAdapter with FlexyMike Dual Ear

https://arstechnica.com/gaming/2020/04/you-can-get-nvidias-rtx-voice-noise-filtering-without-a-pricey-rtx-card/

I followed the link from Ars Technica above and successfully installed the RTX noise filtering software on my laptop which has a GTX.

No license constraints that I can see, but I would not be at all surprised to find out that some future release of the software refuses to run on my GTX laptop. Which may not be too bad - I have been considering replacing my mortally wounded Microsoft Surface Book to with the Surface Book 3 that has an RTX Quadro. In fact, one of the big things holding me pack from purchasing the Surface Book 3 with RTX is my concern that this GPU consumes considerably more power than the GTX alternative, and my experience so far has been that intermittent fan-based cooling on laptops is not good for speech recognition. This is a sort of test drive.

I think it's highly likely that the noise cancellation may work well, but only in steady-state. But there may be problems when the noise level changes, e.g. in the furnace fans turn on or off.

Too early to tell so far, but I thought I would just provide the link above.

Like I said, too early to tell anything so far, except I somehow "feel" that there is a bit more latency in the recognition. With the Dragon parameters set to as fast as possible.

Q: does anyone have a set up to measure latency from saying something, to text appearing? of course the best way would be from an external sound source, but there might be value doing it just inside the software stack.

This latency is different from the occasional long latency delays or stalls that I have seen with Dragon. This latency seems to be rather consistent with the RTX software.

 

---

 

By the way, I saw presentations on this software when I was at Nvidia. It really does look good.  and although it does seem to run on my older GPU, the newer GPUs do have dramatically increased computational bandwidth, which may translate to reduce latency depending on the workload.

 

-------------------------

DPG15.6 (also DPI 15.3) + KB, Sennheiser MB Pro 1 UC ML, BTD 800 dongle, Windows 10 Pro, MS Surface Book 3, Intel Core i7-1065G7 CPU @ 1.3/1.5GHz (4 cores, 8 logical, GPU=NVIDIA Quadro RTX 3000 with Max-Q Design.

I have already mentioned this:

 

DragonBench

I am currently overhauling it. Wait a few days for the next release to be published.

-------------------------

The New Game in Town: DragonConnect

The application has been updated now:

DragonBench

 

-------------------------

The New Game in Town: DragonConnect

Early results:

Nvidia's voice/noise canceling does seem to help a bit.

However, my PC, with an MX250 GPU (Pascal, now quite obsolete, not even a GTX), feels significantly more sluggish.

When the Dragon microphone is on the "NV RTX Voice.exe" consumes 5 to 6% of CPU. Even when nothing is being said. Apparently this is the cost of doing this voice "noise cancellation" on background noise when nothing is being said. Note that this is CPU, not GPU utilization. Process Explorer reports no GPU utilization. Task Manager says that the GPU is being utilized, but the Copy Engine, not the actual SM Streaming Processors. This may make sense, just copying data in and out of the GPU, where it is quickly determined to be quiet.

Even this low utilization makes my PC fans turn on. The constant background noise seems not to hurt speech recognition, but when they ramp up and down, it hurts.

There seem to be more occurrences of the annoying 20+ second stalls. As might be expected, with yet another concurrent process in the mix, plus 10 more threads.

--

Anyway, much of this is probably an artifact of my old MX250 GPU. 12.4 GFlops FP16, 797 GFlops FP32, versus 11.20 TFlops FP16, 5.599 TFlops FP32 on the Quadro RTX 3000 Max-Q in a more recent machine like the Microsoft Surface Book 3 high end. I.e. FP16 = one thousand X faster, FP32 circa 60X faster. I expect that NVIDIAS RTX Voice uses FP16, but I don't know for sure. FP16 on the old Pascal based GPU was not really in hardware, but was provided for SW development, amd was hence slower than FP32; whereas on the newer machines FP16 is circa 2X faster than FP32, as one would expect.

--

BOTTOM LINE: NVIDIA RTX Voice seems to work even on this old machine. It feels like it improves quality, but at the cost of making this older but muscular PC feel sluggish.

I look forward to trying NVIDIA RTX Voice on a modern GPU, with Dragon.




-------------------------

DPG15.6 (also DPI 15.3) + KB, Sennheiser MB Pro 1 UC ML, BTD 800 dongle, Windows 10 Pro, MS Surface Book 3, Intel Core i7-1065G7 CPU @ 1.3/1.5GHz (4 cores, 8 logical, GPU=NVIDIA Quadro RTX 3000 with Max-Q Design.

Trying this while dictating medical progress notes, there is barely perceptible increase in latency, but accuracy seems to be better.

 

Q; what's your configuration? e.g. CPU/GPU/frequency?

 

 like I said, I have a rather old Pascal generation GPU.

-------------------------

DPG15.6 (also DPI 15.3) + KB, Sennheiser MB Pro 1 UC ML, BTD 800 dongle, Windows 10 Pro, MS Surface Book 3, Intel Core i7-1065G7 CPU @ 1.3/1.5GHz (4 cores, 8 logical, GPU=NVIDIA Quadro RTX 3000 with Max-Q Design.

ag,

Memory: 64gb

Microphone Brand and Model: phillips speechmike premium/ANDREA usb pod/KB HANDS FREE MIC- mostly use the old phillips

Operating System: WIN 10 home 64 bit

Processor (CPU) Type & Speed: Intel(R) Core(TM) i7-9700 CPU @ 3.00GHz, 3000 Mhz, 8 Core(s), 8 Logical Processor(s)

Soundcard Brand and Model: nvidia Geforce GTX 1660 Ti

Speech Software (exact version and edition): DPI 15.3

Speech Software Version: 15

@docinfiniti: thanks.  Your system is a lot more powerful than mine, which might explain why you're getting negligible slowdown for NVIDIA RTX voice.

 When next I can afford a hardware upgrade…

 

-------------------------

DPG15.6 (also DPI 15.3) + KB, Sennheiser MB Pro 1 UC ML, BTD 800 dongle, Windows 10 Pro, MS Surface Book 3, Intel Core i7-1065G7 CPU @ 1.3/1.5GHz (4 cores, 8 logical, GPU=NVIDIA Quadro RTX 3000 with Max-Q Design.