Curious about how Dragon NaturallySpeaking work under the hood to utilise hardware. 

It was mentioned in another post that the speed of a single CPU is essential for the performance of Dragon, not multiple cores. And apparently not the GPU either. After version 15, SR in Dragon is based on deep learning algorithms, i.e. deep neural networks. They run significantly faster on GPU's because of GPU's parallelisation which makes all the matrix multiplications and tensor operations in deep learning much faster..

So the question is does Dragon take advantage of the GPU available on a PC.? I have a 4 GB Nvidia graphics card on my laptop but it does not seem like Dragon is using it. Why Not? 

 

Rune Vabø

Intel(R) Core(TM) i7-4710HQ CPU @ 2.50GHz, 2501 Mhz, 4 Core(s), 8 Logical 

16 GB of RAM

256 GB SSD

Nvidia GeForce GTX 860 M, 4 GB

mic: Sennheiser SE630

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

Rune Vabø

According to the Windows 10 Task Manager, Dragon does utilise the GPU.

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

No need to buy if all you want to do is try ...

DragonCapture KB Download (Latest)
DragonCapture Homepage

a little bit off topic, but if you want to use your GPU check out

Nvidia RTX Voice Software Noise Cancelling - OMG

https://www.knowbrainer.com/forums/forum/messageview.cfm?catid=6&threadid=34359&highlight_key=y&keyword1=RTX

 

This has greatly improved my speech recognition experience. Unfortunately you need a pretty recent Nvidia RTX GPU - although there were patches available that allowed it to run on older GPUs. (Why in the world should ML noise filtering software require ray tracing hardware?)

 

but even this noise filtering software uses only a very small amount of GPU horsepower.

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

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.

Hi Ag, when you say it improved your recognition experience, are you talking about 1) increased resistance to noise, 2) increased accuracy, or 3) a noticeable speedup in recognition speed that would show up on Rudiger's Dragon Recognition speed tool?

It might be worth it to buy a new Nvidia graphics card to speed up recognition if the GPU help cut recognition speed in half.

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

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, 3 BenQ 2560x1440 monitors, Microsoft Sculpt Keyboard and fat mouse

Kevin,

First off, you probably meant to say "double the recognition speed" rather than "cut recognition speed in half."

As regards the feasibility of what you are having in mind, please be so kind and have a look at a recent posting of mine, referring to the so-called "real-time factor", which has been a parameter to measure and judge the performance in the R & D of speech recognition since early on.

https://www.knowbrainer.com/forums/forum/messageview.cfm?catid=4&threadid=35807&highlight_key=y#188565

 

 

Now, assuming that I haven't failed at making myself clear enough in the message, would you consider it is even possible to double the speed of Dragon recognition by today's standards?

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

No need to buy if all you want to do is try ...

DragonCapture KB Download (Latest)
DragonCapture Homepage

Hi Rudiger, it was exactly because of your previous posting that anything < 1 (like 0.85) was pretty good. I was going to download your program to see what my number was, but never got around to it.


My current question to Ag was to see if he noticed any improvement in recognition speed. I was thinking if Dragon was CPU bound for recognition processes, and if it could use parallelism, and if it could use GPUs to overcome some of its limitations (you can see how this chain of thinking devolves into meaningless if statements and complexity)... So I just asked Ag if he noticed a faster response time. It seemed simpler. I expect an answer like "No, the GPU didn't help recognition speed at all," but I would like to be surprised with "Yes! I noticed an amazing improvement!" :-)

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

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, 3 BenQ 2560x1440 monitors, Microsoft Sculpt Keyboard and fat mouse

Kevin,

I tend to take "this chain of thinking devolves into meaningless" as the keyword in your previous response.

Basically, there is no way to double the speed of Dragon recognising your speech, no matter how hard you try; on a technical level. However, there are some ways to avoid Dragon falling below its built in capabilities with regards to speed, some of which are avoiding less than optimal input, extraneous noise, and the likes of it. It will always be as simple as that.

If you ever suffer from Dragon not responding instantaneously, these are the first things you should look into.

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

No need to buy if all you want to do is try ...

DragonCapture KB Download (Latest)
DragonCapture Homepage