NAM: Neural Amp Modeler

[Mike Oliphant]

However, as it turns out this hotone jogg has at least 6 dB higher input noise floor than my clarett, and even more so against the tonex as an interface.

Interesting. I'm using a Jogg and I have no issues with noise. I don't play super high gain stuff, though.

I just did a recording test, and I'm seeing a less than -70db noise floor with an instrument connected with the strings muted.

Maybe you have some interference going on? Or are you seeing noise at the output end?

 

[northern_fox]

Interesting. I'm using a Jogg and I have no issues with noise. I don't play super high gain stuff, though.

I just did a recording test, and I'm seeing a less than -70db noise floor with an instrument connected with the strings muted.

Maybe you have some interference going on? Or are you seeing noise at the output end?

I mean it's definitely functional, but with higher gain amps it just seems to have a pretty big noise floor. My test was to reamp the same DI via my signalart through multiple interfaces along with a sine wave test tone for input level matching, and then compare the pre-guitar line noise only portions and the guitar portions between each one. It could be done more scientifically I'm sure, but this was my quick and dirty test. In order of noise floor, I found the following:

ToneX < Axe 3 < clarett < scarlett solo < < jogg

I know it isn't output related noise because I didn't connect or monitor through outputs for any of the above - just one USB connection and one input. I'm not bothering to report absolute noise floor # because as you say, it's relative to a lot of factors. Just looking at the relative differences.

 

[northern_fox]

For the record - measuring the noise level of the silent section _after_ a recto capture (because the normal floor was too low to measure), the lowest I was able to get jogg was -45 dB, where the axe/tonex were around -56 dB, and the best tonex result with tweaking the input levels was -63 dB. Focusrites were around -51 dB in there. Again, just relative difference since and the absolute # is kind of meaningless. Same results if instead of using an amp I just boost the signal itself high enough to measure.

Possibly makes sense why I felt it was so noisy, because I'm used to using axe / tonex when playing via real cabs, and jogg was 10-20dB higher compared to those which is really a lot when the room volume is cranked

 

[Mike Oliphant]

My test was to reamp the same DI via my signalart through multiple interfaces along with a sine wave test tone for input level matching

I assume that means that you are using the sine wave tone to ensure that the digital levels can be sensibly compared? The relevant measure to compare is signal-to-noise ratio (which is the same as just comparing digital dB if the signal levels match).

Obviously, the digital dB level you get from the Jogg will depend on where you've got the gain knob set. I'm running mine at halfway (which my humbuckers can fully saturate at full volume).

 

[northern_fox]

I assume that means that you are using the sine wave tone to ensure that the digital levels can be sensibly compared? The relevant measure to compare is signal-to-noise ratio (which is the same as just comparing digital dB if the signal levels match).

Obviously, the digital dB level you get from the Jogg will depend on where you've got the gain knob set. I'm running mine at halfway (which my humbuckers can fully saturate at full volume).

Sine wave was just to give me a reliable reference point for normalizing the input levels to make them comparable yeah. I run the jogg ~ 10 oclock, which was somewhere around the same input level as gain 0 on my focusrite.

I didn't really set out to make the most scientific test I could - I'm mainly focused on picking what works best for me live based on my normal use case

 

[Mike Oliphant]

Playing around a bit more with this, since it has gotten me curious.

I did another test - this time with an instrument connected, but volume knob at zero. Jogg gain at ~10 o'clock:

Pretty low. But check out the dc offset - there is a significant offset from zero.

If I correct the dc offset, I get this:

Which is *very* good.

I don't think a dc offset like this would be audible, but it may well impact plugins if they aren't designed to handle it. And it obviously impacts the RMS dB calculation.

 

[northern_fox]

Playing around a bit more with this, since it has gotten me curious.

I did another test - this time with an instrument connected, but volume knob at zero. Jogg gain at ~10 o'clock:

View attachment 7561

Pretty low. But check out the dc offset - there is a significant offset from zero.

If I correct the dc offset, I get this:

View attachment 7562

Which is *very* good.

I don't think a dc offset like this would be audible, but it may well impact plugins if they aren't designed to handle it. And it obviously impacts the RMS dB calculation.

What are you measuring with here? My DAW tools aren't apparently low enough to resolve this stuff by itself and I'd love to be able to do more accurate measurements.

I did just order an audient evo 4 to see how that measures up. Avoiding the xsonic because I don't need any buttons. At the end of the day, I care most about what actually works best in the room and noise isn't a huge dealbreaker, but I'd still like to use what will be the best compromise of tone/noise/size. The point I kind of consider it to be unusable at is when it gets noisier than an actual amp/cab. This is all for live/loud in my case. At home it's all good.

 

[Mike Oliphant]

What are you measuring with here?

I'm using Audacity: "Analyze->Measure RMS"

Btw, I experimented with dc offset correction (you just need to run a really low high-pass filter) in front of a high-gain model in NAM, and it didn't have any audible impact that I could discern. It may be worth trying with your signal chain, though.

As you say, it is what works at the end of the day. It is certainly possible that other interfaces have a 10-20 dB lower noise floor. If your amp model is pushing enough gain to shift -120dB into the problematically audible range, then reducing that by 10-20 dB will make a real improvement. Or you could just use a noise gate...

 

[northern_fox]

I'm using Audacity: "Analyze->Measure RMS"

Btw, I experimented with dc offset correction (you just need to run a really low high-pass filter) in front of a high-gain model in NAM, and it didn't have any audible impact that I could discern. It may be worth trying with your signal chain, though.

As you say, it is what works at the end of the day. It is certainly possible that other interfaces have a 10-20 dB lower noise floor. If your amp model is pushing enough gain to shift -120dB into the problematically audible range, then reducing that by 10-20 dB will make a real improvement. Or you could just use a noise gate...

Oh I do use a gate definitely - just trying to minimize it and the tonex rig was overall a lot quieter than what I was getting with joggtop. I'm not done trying yet though.
I can use the reaper analysis similarly:

Jogg:

Axe 3:

ToneX:

Clarett:

 

[northern_fox]

The evo 4 might be the perfect balance here for the same money. Will find out this weekend after it shows up

 

[Mike Oliphant]

I can use the reaper analysis similarly:

Jogg:

I'd be curious what those RMS numbers look like after dc offset fixing. If your Jogg is like mine, the -90dB will go down significantly. Curious about whether that's the case for the other devices as well.

 

[northern_fox]

I'd be curious what those RMS numbers look like after dc offset fixing. If your Jogg is like mine, the -90dB will go down significantly. Curious about whether that's the case for the other devices as well.

What do you mean by DC offset? is this a physical correction you're making?

 

[Mike Oliphant]

What do you mean by DC offset? is this a physical correction you're making?

Audio with a DC offset means that the waveform is not centered on zero, but is offset.

Look at the images I posted above. See how the first one is centered above zero? That artificially inflates the noise floor calculation because it is based on deviations from zero.

The easiest way to get rid of it is to do a high-pass filter at as low a frequency as you can. In Reaper, load up ReaEQ, switch the "Type" of the leftmost point to "High Pass", and drag it all the way to the left.

You'll have to render this out to a .wav file and then load it back in as a new item if you want to use the Reaper item analysis to get RMS dB.

 

[northern_fox]

Audio with a DC offset means that the waveform is not centered on zero, but is offset.

Look at the images I posted above. See how the first one is centered above zero? That artificially inflates the noise floor calculation because it is based on deviations from zero.

The easiest way to get rid of it is to do a high-pass filter at as low a frequency as you can. In Reaper, load up ReaEQ, switch the "Type" of the leftmost point to "High Pass", and drag it all the way to the left.

You'll have to render this out to a .wav file and then load it back in as a new item if you want to use the Reaper item analysis to get RMS dB.

interesting, that reduced it by 8dB to -103. How exactly is that working to remove the offset? Can I just apply a similar filter before NAM chain to do the same in realtime?

 

[Mike Oliphant]

interesting, that reduced it by 8dB to -103. How exactly is that working to remove the offset? Can I just apply a similar filter before NAM chain to do the same in realtime?

The dc offset effectively has a frequency of 0, so it gets removed by any high pass (low frequency removal) filter.

And yes, you can just run it at the front of your chain.

 

[northern_fox]

Ok - I may be partially (entirely) to blame here. I've been testing everything via reamp - so out of my main PC, through reamp, into the other interface. It seems like the noise level (which was really, very audible even during non silent parts) has dropped amount when I just plug in a guitar and have it on silent or even open volume. So I think it may be more usable than I realized - it just doesn't work that well to reamp into one interface from another in this instance, for whatever reason. While it _is_ still noisier than the others, I think you're right and it's not a big enough difference to matter. That floor was just magnified a lot overall by some ground loop or other weirdness between these devices. Still looking forward to giving the Evo a shot though.

 

[aivoryuk]

as this thread seems to contain a lot of comments about testing then you may want to check out the latest Dan Worrall video

 

[MirrorProfiles]

https://samply.app/p/jLo8NkVD5Gla1udh1VLK

its a Rev G Dual Rectifier, captured with a cabinet as a load. One of the clips is the real amp, one is a NAM model, one is Helix, one is STL Amphub. If you like, make a guess as to which is the real amp and which is NAM, and then PM me and I'll share a capture of the orange and red channels.

650 Epochs. 0.011 ESR.

 

[northern_fox]

https://samply.app/p/jLo8NkVD5Gla1udh1VLK

its a Rev G Dual Rectifier, captured with a cabinet as a load. One of the clips is the real amp, one is a NAM model, one is Helix, one is STL Amphub. If you like, make a guess as to which is the real amp and which is NAM, and then PM me and I'll share a capture of the orange and red channels.

650 Epochs. 0.011 ESR.

A - STL
B - NAM
C - Real amp
D - Helix

Didn't look at waveforms or DAW tools
They sound so freaking good though - all of them.

 

[MirrorProfiles]

A - STL
B - NAM
C - Real amp
D - Helix

Didn't look at waveforms or DAW tools
They sound so freaking good though - all of them.

Ha I may as well reveal now (I think only you and @Deadpan guessed). I'm not sure I even like the tone now, I'm pretty sure I just used the first IR that came up in my default Libra patch (which is a greenback mic'd dead centre - something I'd never use on a rectifier).

B was NAM, C real amp. A Helix, D STL. What surprised me most was I dialled this in in the room, recorded the line out from the amp while the cab was connected. Didn't even adjust the tone to how it sounded recorded. But the sound I dialled in the room was basically the sound I have saved as my default in Helix give or take. Kind of nuts how our ears just push us in a similar direction regardless of the gear being used.

PM me @northern_fox if you'd like the NAM files (I know you have your own Recto and probably tons of models made already)