Jump to content

Let's compare real-time noise reduction algorithms [sound clips included]


ninjafreddan
 Share

Recommended Posts

Hi

 

I'm using a Cedar DNS2 noise reduction unit whenever I have the need for it in the field.

 

It would be interesting to compare it with other real-time algorithms like the Sound Devices NoiseAssist and Cedar sdnx plug-ins. And if there are other solutions available, why not compare them as well.

 

I've recorded a couple of sound bites with my shotgun mic and processed them through the Cedar DNS2.

 

If anyone on the forum can run the same unprocessed sound bites through a MixPre-10 using NoiseAssist set to the same amount of noise reduction, that would be awesome. Same goes for the Cedar sdnx plug-in in the 8-series mixers. Preferably via AES/EBU whenever possible to keep out other factors like the sound of the converters etcetera.

 

I've used the following reduction values: -3, -6, -9 and -20dB

 

The [Learn]-function was enabled, and I believe it's always on in the Cedar sdnx plug-in, so they should have similar performance - but let's compare them.

 

Mind my recorded English, and if anyone want to add troublesome recordings that need noise reduction, please feel free to do so, and I'll run them through the Cedar DNS2.

 

Naturally, it would be interesting to compare the results with offline processing from Izotope, Waves etcetera, but let's keep it to real-time noise reduction in this thread, okay. Stuff that can be helpful in the bag or in the cart.

 

 

Thanks in advance

Frederick

 

01_-_Voice_recording_unprocessed_48kHz.wav 02_-_Voice_recording_with_Ceder_DNS2_3dB_noise_reduction_48kHz.wav 03_-_Voice_recording_with_Ceder_DNS2_6dB_noise_reduction_48kHz.wav 04_-_Voice_recording_with_Ceder_DNS2_9dB_noise_reduction_48kHz.wav 05_-_Voice_recording_with_Ceder_DNS2_20dB_noise_reduction_48kHz.wav

Link to comment
Share on other sites

Thanks Frederick, I'm very grateful that you've submitted these.

I've uploaded the difference in the conditioned recording and the original signal from the Ceder files.

 

I would prefer that you, Constantin, NOT comment nor state your opinion on this test, as you have already publicly and unsolicitedly expressed "I am not interested in this particular test". And to be clear, I am not seeking an indication of your interests nor opinion on the matter. You are entirely relieved of the job that compels you to comment here. And I would prefer that this contribution remain constructive, uphold the author's topic, and serve only those participants in our community who are interested in the results. 

 

I don't have noise assist for the mixpre, and I am evaluating whether to make the purchase. I hope other members will take on your request and translate the test to the Sound Devices units.

 

These files were saved without dither, as they were not converted from the original 24bit, 48K files. An offset of either .02ms or 1 sample placed on the original file resulted in the most complete polarity at this level of precision.

 

Voice_recording_with_Ceder_DNS2_3dB_NoiseOnly.wav Voice_recording_with_Ceder_DNS2_20dB_NoiseOnly.wav Voice_recording_with_Ceder_DNS2_9dB_NoiseOnly.wav Voice_recording_with_Ceder_DNS2_6dB_NoiseOnly.wav

Link to comment
Share on other sites

11 hours ago, James Louis said:

You are entirely relieved of the job that compels you to comment here


Well, it’s a good thing you don’t get to decide that. 
in any case, I did say that an AB test might be interesting, so I‘m not even sure what you’re going on about 

Link to comment
Share on other sites

On 4/12/2021 at 3:26 AM, Jebari said:

Don't have any of the SD mixers anymore, but thought it might be interesting to throw in a Zaxcom Nova to the mix, even though I may be wrong but I believe its NR is more of a noise gate than the Cedar and SD style real time NR plugins.

 

Thanks for the sound clips. I've never used a Zaxcom mixer so I didn't know that it had that kind of processing.

 

Like you point out, it sounds more like a noise gate, but even if it has a Threshold setting in the mixer, it sounds like the Threshold is adaptive - because it performs better than a crude noise gate roughly set to the same settings. I think it performs pretty well and it doesn't make the ambience sound 'processed'.

 

With a couple of lavs open and maybe a boom, you can probably get away with it sounding a bit jumpy now and then, right? Cleaning up the camera feed a bit.

 

How are you using it in the field?

 

 

Cheers

Frederick

Link to comment
Share on other sites

8 hours ago, ninjafreddan said:

 

Thanks for the sound clips. I've never used a Zaxcom mixer so I didn't know that it had that kind of processing.

 

Like you point out, it sounds more like a noise gate, but even if it has a Threshold setting in the mixer, it sounds like the Threshold is adaptive - because it performs better than a crude noise gate roughly set to the same settings. I think it performs pretty well and it doesn't make the ambience sound 'processed'.

 

With a couple of lavs open and maybe a boom, you can probably get away with it sounding a bit jumpy now and then, right? Cleaning up the camera feed a bit.

 

How are you using it in the field?

 

 

Cheers

Frederick

Yeah totally, it actually worked surprisingly well to cut down the reverb tails without sounding overly processed. They have it built into the auto mixer (though you can set it to NR only) and I've used it on 3 - 6db along with automix when I have a handful of lavs out, but honestly haven't used it heavily yet. This test actually got me more interested in trying it in live rooms though to clean up the mix track a bit.

Link to comment
Share on other sites

13 hours ago, Jebari said:

Yeah totally, it actually worked surprisingly well to cut down the reverb tails without sounding overly processed. They have it built into the auto mixer (though you can set it to NR only) and I've used it on 3 - 6db along with automix when I have a handful of lavs out, but honestly haven't used it heavily yet. This test actually got me more interested in trying it in live rooms though to clean up the mix track a bit.

 

I think it sounds fine and if you have two-three channels open, you're not going to notice the gate working on each channel - it will be masked by the ambience, and clear things up a bit.

 

The difference with proper noise reduction - like the Cedar DNS2 - is that I can apply it on the camera feed and set the noise reduction to 5dB, and it just work wonders - taking out room acoustics and troublesome ambiences dynamically both in level and in frequency range.

 

But in your case, noise gate and automix on each channel will be almost as effective - and it's built into your Nova mixer.

 

 

Cheers

Frederick

Link to comment
Share on other sites

I have one on my cart and I use a 688. I am still learning how to use it. I have both attenuation and bias at about -4. I have read that it will work better with AES in and out, but all six of my inputs are designated to my wireless. I send the mix to x3 and x4 and directly to the Cedar then out put to channel 8 and 9. have left and right output turned off and record them pre faded. 

Link to comment
Share on other sites

On 4/19/2021 at 12:47 AM, newzhack said:

I have one on my cart and I use a 688. I am still learning how to use it. I have both attenuation and bias at about -4. I have read that it will work better with AES in and out, but all six of my inputs are designated to my wireless.

 

That sounds like a good start. I did experiment with the bias setting a bit when I first got the unit, but eventually ended up at bias 0, and just adjust the attenuation according to the situation. But, it's great to be able to lower or raise the bias-setting.

 

I'm mostly using the digital I/O, but now and then use the analogue - and I think it sounds great.

 

On 4/19/2021 at 12:47 AM, newzhack said:

 

I send the mix to x3 and x4 and directly to the Cedar then out put to channel 8 and 9. have left and right output turned off and record them pre faded. 

 

If you have a spare stereo AES/EBU output on the 688, then you can at least send a digital signal into the Cedar DNS2. You have four channel AES/EBU outputs on the 10-pin A output - fancy some soldering? :-)

 

Digital inputs are assigned to channel one and six, I guess, like on my 664.

 

You probably already know that the analogue part of the DNS2 is powered down when digital I/O is selected - a power-safe feature - but the analogue side can be enabled by pressing [SETUP] and then both [DNS ON]-buttons at the same time. I believe it has to been done each time the unit is powered on.

I sometimes use it by sending Digital Out from my 664, and then analogue out from the DNS2 to a headphones distribution and the digital out to a recorder - when I feel like spoiling my client. Usually end up at about 4dB noise-reduction, like you. Cedar noise-reduction is a no-brainer in those occations.

 

I always use it for events and conferences, as a digital insert in my mixing console, channel one noise-reducing a mix of the head mics, and channel two the podium mics. Having the ability to noise-reduce the podium mics by 6dB, means that I can handle the most shy presenter without any problem at all - that alone is worth the price of the unit - for me. :-)

 

It would be great if anyone on the forum could process my sound bites through an 8-series Sound Devices, so that we can compare the results. Thanks in advance!

 

 

Cheers

Frederick

Link to comment
Share on other sites

  • 3 weeks later...

This is the noise output only of the Nova. This was a trickier experiment to conduct. If you would also like your equipment or parameters to be tested as such, confirm that you are processing the entire original file, perhaps by starting recording on playback, rather than rolling independently. Keep your faders at unity and confirm that there is not any global output compensation on the noise reduction. Also, post the unprocessed file along with it and confirm that you haven't changed the length of each file. If the unit has a bounce or export option, that's the best way to confirm that the files will line up. I had to resample the files rather than export the results in this case, which explains the odd reference numbers in the file names. I apologize that each example isn't more immediately clear.

 

That said, this experiment was done with a 3 sample offset from the original, after matching the exact sample by scrubbing the conditioned audio in place (sample offset corresponds to longer processing time on the Nova as well as where there is the most complete polarity in order to conduct this experiment.). Then the files were gain matched, which is in itself an oxymoron, as there is an attempt in the algorithm to reduce the noise output by the corresponding amount of attenuation. You can confirm that the noise reduction is working against the original file on your meters in the sections of the recording where the Nova is listening only to the ambient noise.

 

That said, the Nova still appears to be attenuating the entire file considerably more than the Cedar in the output of this experiment, however I wanted to focus on only the information that was either retained or lost. If you are hearing something in this file, that's what you've lost. If I had not gain matched the files, it would further highlight the volume drop of the treated version. In this case, you will hear that much more signal is lost on the Nova than with the Cedar. This is a good illustration of the trade-offs between nuanced algorithmic noise reduction and noise gating.

500981753_noiseoutputonlyNovaNR-3.aif 1762583870_noiseoutputonlyNovaNR-6.aif 1011908702_noiseoutputonlyNovaNR-9.aif 1880584940_noiseoutputonlyNovaNR-20.aif

Link to comment
Share on other sites

On 5/12/2021 at 3:47 PM, James Louis said:

That said, the Nova still appears to be attenuating the entire file considerably more than the Cedar in the output of this experiment, however I wanted to focus on only the information that was either retained or lost. If you are hearing something in this file, that's what you've lost. If I had not gain matched the files, it would further highlight the volume drop of the treated version. In this case, you will hear that much more signal is lost on the Nova than with the Cedar. This is a good illustration of the trade-offs between nuanced algorithmic noise reduction and noise gating.

 

Hi James

 

Thanks for your noise component comparisons. They work for my Cedar-processed files because I put the original recording in Cubase Pro, added an External FX plug-in and inserted the Cedar DNS-2 via AES/EBU - so every clip has the same timing and level.

 

But, the Nova noise residue files doesn't sound as they should. There should be more pumping action as the noise gate work the material, more noise-to-signal action.

 

Did you use Jebari's sound clips, or have you made your own sound clips in a Nova?

 

It would be great to have level-matched Nova processed files using the AES/EBU-input of the machine. Maybe add a short tone in the beginning of the original sound clip as a level and timing reference. I should have thought of that when I created my files.

 

When I have the time, I will update my files.

 

 

Have a nice weekend

Frederick

Link to comment
Share on other sites

Hi Frederick, I don't have access to a Nova. I am using the clips posted on here. So any deviation is a matter of what was provided.

 

I would bring to your attention the underlying hunch that you would hear more pumping. This is not necessarily the case. That's what we're trying to discern. It could be that the release on a gate, for example, is gradual enough that its not grabbing our attention, yes, but it could also mean that the Nova is using noise reduction, ie filtering. Although it is very tricky to subsequently volume match the original, I believe that you can still conclude something by confirming how much of the signal (rather than the noise) is targeted in the processing.

 

Even if the original were louder, I can still claim with a listen that the range of the information in the signal is passed into the processing. Even if I only reached partial polarity, I can still partially evaluate what of the spectrum is differentiated in the test. You can confirm this by measuring the -dB of just the noise bed or putting it back into a spectrogram, and you'll see that my volume leveling isn't too far off, and I did measure from a tenth of a decibel. From what is conveyed in listening to the content, not just the context, there is quite a bit of discernible dialog lost here which I cannot otherwise explain, and I'd be happy to be wrong if it means our Nova users are getting more from their equipment.

 

Still, yes, this would be more controlled experiment if we post the clips in the manner you instructed!

Link to comment
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

 Share

×
×
  • Create New...