nvidia Issues with Nvidia Broadcast and Sampling Rate

[TechSwagPimp1337]

If anyone is having issues with distorted audio while using Nvidia's noise removal for either microphone or speakers, I've found that the issue for me is using 192kHz sampling rates causes issues. I believe the program only supports 44.1kHz, 48kHz, 16-bit, and 24-bit. Kind of strange because more data = better performance, and the RTX cards can absolutely filter the 1x1024 signal every 2ms. Even a $100 DAQ can do complex Fourier mathematics that fast. RTX cards should support it.

[Z_G]

Yes, after building a new computer I am experiencing the same issue. Going above 96kHz 24-bit on playback completely breaks the application both recording and playback.

[The Flying Sloth]

Going above 44khz has no benefits for audio quality so I really don't think this is an issue.....

[Derkoli]

On 11/30/2020 at 2:56 PM, TechSwagPimp1337 said:

If anyone is having issues with distorted audio while using Nvidia's noise removal for either microphone or speakers, I've found that the issue for me is using 192kHz sampling rates causes issues. I believe the program only supports 44.1kHz, 48kHz, 16-bit, and 24-bit. Kind of strange because more data = better performance, and the RTX cards can absolutely filter the 1x1024 signal every 2ms. Even a $100 DAQ can do complex Fourier mathematics that fast. RTX cards should support it.

As @The Flying Slothsaid, just used 44.1kHz/16bit. Most stuff is done in 44.1kHz/16 bit now-a-days, as it's just an ideal sample rate/bit depth to nearly maximise dynamic range from the very best amplifiers and get a maximum HF response of 22.05kHz, Which we can't hear under normal conditions.

 

I've done some audio stuff for large, international companies, and it's exported in 44.1kHz WAV. Nothing more, nothing less.

[Z_G]

4 hours ago, Derkoli said:

I've done some audio stuff for large, international companies, and it's exported in 44.1kHz WAV. Nothing more, nothing less.

Maybe at your studio, we mastered everything at 48kHz and others go even higher.

[The Flying Sloth]

50 minutes ago, Z_G said:

Maybe at your studio, we mastered everything at 48kHz and others go even higher.

Don't get me wrong, I record at 96 just for flexibility but every major studio I've spent time in and/or for renders at 44.1

[Ahoy Hoy]

Most live sound systems which use digital conversions do it at 44KHz or 48KHz. Some systems have a digital to analogue conversion multiple times from source to output. I've seen a system do eight conversions from analogue to digital and back from source to output and at no point has anyone ever gone bit rates a bit low losing that clarity in the sound. Because the simple answer is we passed the stage of giving a flying toss thousands of herts ago and by the time it gets played theres crappier items doing worse things to the sound than the rate of conversions.

Nothing wrong with recording in incredibly high bit rates so you have as much data to work with when editing the audio. But that's only suitable in studio environments with perfect recording environments. If you've got to turn on software to remove background noise your not in a studio your in a crappy environment where recording in stupidly high res doesn't matter. And that is why Nvidia didnt waste their time developing the system to work at super high resolution because if you need this software your not using it to do studio recordings your using it so your friends dont hear you mum hovering on discord or for streaming and those software are compressing your audio so small it don't matter when it gets to the end listener.

[Derkoli]

9 hours ago, Z_G said:

Maybe at your studio, we mastered everything at 48kHz and others go even higher.

I tend to master stuff at 192kHz, just for flexibility. But it all gets exported in 44.1kHz or WAV.

[TechSwagPimp1337]

On 12/2/2020 at 7:31 PM, The Flying Sloth said:

Going above 44khz has no benefits for audio quality so I really don't think this is an issue.....

This isn't a simple answer. Recording at high bit-rates actually does provide a benefit to clarity in low-end frequencies. Just because we can't hear frequencies above 22kHz doesn't mean that they don't exist, just like with our eyes not being able to see infrared or ultraviolet waves. Because frequencies higher than 22kHz exist and get picked up by audio recording equipment, it muddies up the low-end quite a bit. The solution is to record at a high sampling rate (96 or 192kHz), apply low-pass filtering, and then render at a lower sampling rate such as 44.1 or 48kHz.

[The Flying Sloth]

39 minutes ago, TechSwagPimp1337 said:

This isn't a simple answer. Recording at high bit-rates actually does provide a benefit to clarity in low-end frequencies. Just because we can't hear frequencies above 22kHz doesn't mean that they don't exist, just like with our eyes not being able to see infrared or ultraviolet waves. Because frequencies higher than 22kHz exist and get picked up by audio recording equipment, it muddies up the low-end quite a bit. The solution is to record at a high sampling rate (96 or 192kHz), apply low-pass filtering, and then render at a lower sampling rate such as 44.1 or 48kHz.

Yeah, if we're going to start getting into nyquist theory you need to understand how little that matters compared to the sound quality degradation of noise removal..... 

[TechSwagPimp1337]

On 12/10/2020 at 6:44 PM, The Flying Sloth said:

Yeah, if we're going to start getting into nyquist theory you need to understand how little that matters compared to the sound quality degradation of noise removal..... 

It isn't nyquist theory. This is research that I've performed in graduate school. Higher sampling rates allow for more information to pulled out of each sample envelope, and therefore better feature extraction related to noise, and better noise removal. I've done extensive research on optimizing real time high-sampling data capture and processing on both audio and biomedical applications.

[The Flying Sloth]

2 hours ago, TechSwagPimp1337 said:

It isn't nyquist theory. This is research that I've performed in graduate school. Higher sampling rates allow for more information to pulled out of each sample envelope, and therefore better feature extraction related to noise, and better noise removal. I've done extensive research on optimizing real time high-sampling data capture and processing on both audio and biomedical applications.

Well no, you're talking about aliasing which (because of nyquist theory) is exactly why we sample at over double the range of human hearing and also why most decent audio platforms cut off frequencies above 20k... You're arguing that aliasing (which should not be present) is a large enough detriment to warrant noise removal. Not only that but you want to do broad spectrum noise removal in real time because you seem so scared of it.... 

 

Unless you're in a studio setting aliasing is a non-issue. The tiny amount of difference it makes just isn't worth the trouble for Youtube/Streaming and if you're really worried, just run a low cut filter and be done with it . There's no point running at a higher sample rate in the hopes of noise removal being higher quality than just running a low pass filter.

 

If you need to do noise removal after recording audio in a studio setting you recorded wrong. But for aliasing on Youtube, an extremely heavily compressed audio signal distribution network, forget it, it's a drop in the bucket.