The No.1 Website for Pro Audio
So I Compared Some Tape Machines and Emulations (No Blind Tests, just Data!)
Old 14th April 2017
  #1
Here for the gear
 
So I Compared Some Tape Machines and Emulations (No Blind Tests, just Data!)

Hi! I’m a very long time lurker of this community but never really bothered to register, since I didn’t feel like I had much to add to the varied discussion in this forum. Now I maybe do have something to add, so hello everybody! A couple of months back I got my Bachelor’s degree in Music Technology. I wrote a thesis on a subject that’s been quite controversial here: how close are tape emulations to the real deal? The thesis is in my native language (Finnish) so I won’t even bother linking it here. However I thought it might be interesting for you guys if I more or less briefly went over the results of my study in this thread.

I’m very interested in this topic myself, since I’m a plugin hoarder and a strong advocate for doing everything ITB. Even my guitar setup live and in the studio is fully digital and modelling. With this in mind, I thought it was time to try and take a deeper, more critical look at some tape emulations to maybe broaden my worldview a little bit. That’s why I pitted my trusty emulations (and one that I’m interested in buying) against some well-kept tape machines I could get my hands on. The participants in the study were: Slate Digital Virtual Tape Machines, Toneboosters ReelBus, U-he Satin with tape machines: Otari MTR-12 and Otari MX-5050Bii.

The Study

I know there’s been a lot of blind testing comparisons between tape machines and their emulations. In my humble opinion, there’s a glaring problem with these kinds of comparisons - they rely entirely on the test subjects’ subjective perception of sound. There are many studies showing that the ears aren’t exactly precision instruments and there are a lot of variables ranging from the quality of the listening environment to the psychological states of the test subjects. I wanted to gather data for my study, not entirely unlike the analysis methods that the actual plugin developers use themselves to gather data for their algorithm needs. Since I do not develop plug-ins, I kind of had to go in blind and make it up as I went along, but I still think that I managed to find some meaningful data and even managed to achieve some conclusions that you might agree or disagree with here.

I used the program Fuzzmeasure to obtain graphs of frequency response, phase response, and THD at different input levels. Surprisingly, different input levels didn’t have much effect on the graphs, so 0dbVU impulses are what I’m going to show you here. Additionally, I prepared a bunch of different audio tracks in Pro Tools and sent them through the processors with, again, varied input levels (-3dBVU, 0dBVU, +2dBVU, +4dBVU and +5dBVU). In the test I included isolated bass drum and snare drum hits, an isolated bass track and a full spectrum mono mix with heavy transients. (Crosstalk was a territory I didn’t want to get into here, so everything was in mono).

For settings on the plug-ins I used the settings recommended by each plug-in developer. This included reading carefully through the manuals of each plug-in and finding the settings (and in some cases presets) that - in the developers view - corresponded with a well maintained 1/4’’ 2-track tape machine, 15ips speed, a modern tape model with plenty of headroom (such as the SM900) and a bias of the most neutral nature in terms of the frequency response. Both of the Otaris also hit these parameters. Of course I’m well aware that all the emulations have modelled different tape machines with different specs from the tape machines I compared them with but I was hoping that by having multiple emulations and multiple tape machines I could find some kinds of trends between the two. I tried to really ask around my area to find a real 2-track Studer for testing but even the pro studios that I contacted had sold theirs or had one that was fallen into a state of disrepair. I included the Pro Tools sessions at the bottom of this post so we could maybe expand on this together if there's interest for that sort of thing!

Some additional technical details: I used an RME Babyface (the original one) for gathering the data. With the tape machines I used the analog inputs and outputs compensating for the input’s frequency response, either by using Fuzzmeasure’s Automatic correction or Totalmix’s internal EQ. With the emulations, I used ADAT to connect digitally to another computer with an M-audio Profire 2626 interface (in slave mode) running the plug-ins in Logic (since Reelbus is AU only). The sample rate for everything was 44.1kHz since that still is what I, my colleagues and most of you guys use in your productions. All of the plug-ins oversample internally.

The Graphs



The Frequency Response

Comparing the frequency responses, I’d say they look pretty similar, except for Reelbus! (VTM’s graph is a bit lower because these are normalized curves and the exaggerated head bump is bringing the whole graph down) All the curves feature the head bump in the lower frequencies, a relatively flat midrange and a slight boost before the drop-off in the high end with the Reelbus acting strange on the high frequencies. As you can see, not even the two tape machines have their head bumps in the same frequencies (the MTR-12 has it’s head bump almost as low as VTM approaching 20Hz while the MX-5050 has a head bump near the 30Hz area.) With Satin you might lose some bass, and Reelbus really eats into the high end with the settings that I did the measurements. See, the best part with the emulations, is that their parameters are really adjustable: Want to tame the VTM’s ridiculous low end bump? No worries, just use the Bass Alignment control! Want to compensate for Reelbus’ weird high frequencies? Just use the Color Adjustment and Low-High emphasis controls. Want more bass with Satin? Increase the head bump. I bet I could’ve gotten the frequency plots resembling each other quite closely with a bit of tweaking. However, this was out of the scope of my study and I found it interesting to find out what the developers themselves consider as “authentic”.



The Phase Response

One reason tape sounds so nice is the way it “delays” low frequencies ie. turns their phase. This results in the warmth and roundness of the low end. This behavior is quite common in the analog domain since transformers and other components tend to do this sort of thing to the phase response. This behavior has also been successfully emulated in all of the plug-ins. The VTM’s graph looks a bit weird under the audible range, but I’m going to hazard a guess that the lowest frequencies get delayed so much that the graph can’t handle it. I don’t know if it’s really relevant or not. The Reelbus is again behaving weirdly on the high end. I guess this might be related to the weirdness of the frequency spectrum, too!



Harmonic Distortion

Ok, here’s where things get a little weird. The graphs vary wildly when it comes to harmonic distortion! The thicker line is the second harmonic, and the narrower line is the third harmonic. The two Otari’s resemble each other close enough but I really can’t say what’s going on with the emulations. However, I could verify that raising input level mostly affected the odd numbered harmonics in every tape machine and emulation, a trait that has been traditionally attributed to tape. This is the part of my study where I definitely felt like I was reaching a bit. I do not know if measuring Harmonic Distortion this way is even relevant to the sound of the tape or if this percentage type-graph was the best way to represent it. I put it here for completeness sake, maybe some of you might have some insight for it.

Test impulses and graphs are all fun and games, but isn’t the goal of these processors to make music? Let’s move on to the second part of my study. The Pro Tools sessions.

The Kicker

OK, let’s firstly get my subjective experience out of the way:

Listening in many different monitoring environments - subjectively - to my ears the difference between the two tape machines are of similar magnitude than the difference between one tape machine and one emulation. I could not reliably distinguish the real tape machines from the emulations while only using my ears!

Now we can try to focus on some quantifiable data by examining the waveforms closely. When I was examining the transient of the snare drum I came across this, see if you can spot it! The order for the tracks is: the original digital recording, Otari MX-5050Bii, Otari MTR-12, VTM, Satin, Reelbus:



The first transient of the snare drum is significantly “rounded” when sent through the actual tape. I tried to find this same type of transient rounding in the isolated electric bass guitar track, but could not find that sort of behavior: Since the transients were softer, the relation between the transient and the rest of the signal remained the same. Whenever there’s a fast overwhelming transient for example a drum hit, the tape tends to round this transient in a way the emulations do not!

I tried to think of a way to investigate this further when I remembered Bob Katz’s definition of crest factor (the difference between peak and RMS levels of the signal). I proceeded to analyze the dynamic range of the full spectrum mix that I had sent through the tape with different input levels: I took sample accurate selections of the clips of differing input levels and used the audio suite Gain plug-in to take the measurements of peak and RMS levels of said clips. I used the values to count the crest factor of my mix for each plug-in and tape machine across all the input levels that I used. Here’s the resulting table:



As you can see. The hotter you send the signal through the actual tape machines, the more they compress the dynamic range. The difference between -3dBVU and +5dBVU is a whopping 1,5dB in dynamic range! The difference with the emulations did not go over 0,3dB and most of that can even be attributed to the Gain plug-in’s rounding errors (see, even the original digital clip varies within 0,1dB). I know there might be a better way to analyze the signal’s transients relation to the rest of the signal but when it shows even by comparing 30 second clips with the crude Gain -plugin, I can only guess that the differences would be more apparent with more advanced tools.

Ok so what causes real tape to handle the dynamic range differently? It’s certainly not compression. Nor is it limiting. It seems more related to the difference in level over the domain of time - almost like the tape can’t really keep up with the faster high level transients - like the travel from the bottom of the waveform to the top takes too long for the tape while “drawing” the waveform and it realizes it needs to start heading back down already. I wouldn’t really know how to explain it better and it might even be that I’m way off with my analysis, but that’s why I’m also interested in what you guys have to say of all this. Can this type of behavior even be modelled? I don’t know, that’s really a question for the really people behind the algorithms!

We also need to ask how important is this reduction in the dynamic range of an audio signal while reaching for the authentic sound of tape. I think it boils down to what kind of aesthetic you’re going for. If you’re going for a vintage flavored, distorted track you might actually get by very well with an emulation, since they’ve nailed the frequency response and add the same flavor of harmonics when driven hard. However, I personally think that transient control is very important in all sorts of rhythm based music and I think that this weird rounding of the transients really helps and guides the engineer to make pleasant sounding and weighty recordings that are traditionally associated with the sound of tape.

If anything, I hope this thread maybe raises non flammable discussion on the topic, maybe even some solutions how we could capture this kind of transient response digitally, either by improving the algorithms some way or maybe with some completely other tools (such as a certain type of saturation, etc.) Thanks for bearing with me this far. English is not my first language, and I hope it doesn’t show too much. I also realize this is one hell of a first post to make and that’s why I’m going to get a beer now, cheers!

EDIT: Here's the Pro Tools file
Here's the the link to the test session if you want to take a closer look and listen! The samples for the drum-part of the test were performed by the awesome Jari Salminen of Poets of the Fall!
Old 14th April 2017
  #2
qwe
Lives for gear
 
Erm, there are no messages on this page...?!
Old 14th April 2017 | Show parent
  #3
Here for the gear
 
The thread is bugged somehow... I'm gonna wait for the support to handle it. I promise, it's gonna be awesome!
Old 14th April 2017 | Show parent
  #4
qwe
Lives for gear
 
Quote:
Originally Posted by Villberg View Post
The thread is bugged somehow...
Old 15th April 2017 | Show parent
  #5
Lives for gear
 
🎧 15 years
There's part of this I don't agree with. Like the two phase reverse switches in series-
Attached Thumbnails
So I Compared Some Tape Machines and Emulations (No Blind Tests, just Data!)-dawvstape.png  

Last edited by thenoodle; 15th April 2017 at 06:38 PM..
Old 19th April 2017
  #6
Here for the gear
 
Ok, the original post seems to be back up! Talk about a great first impression!
Anyways - enjoy and tell me what you think!
Old 19th April 2017
  #7
qwe
Lives for gear
 
Quote:
Originally Posted by Villberg View Post
Ok so what causes real tape to handle the dynamic range differently? It’s certainly not compression. Nor is it limiting. It seems more related to the difference in level over the domain of time - almost like the tape can’t really keep up with the faster high level transients - like the travel from the bottom of the waveform to the top takes too long for the tape while “drawing” the waveform and it realizes it needs to start heading back down already. I wouldn’t really know how to explain it better and it might even be that I’m way off with my analysis, but that’s why I’m also interested in what you guys have to say of all this. Can this type of behavior even be modelled? I don’t know, that’s really a question for the really people behind the algorithms!
See this post by Paul Frindle...

Also to quote from Urs Heckmann:

"With basic this and basic that one can not create a model which has interacting parameters. What we have is a complete enough model that has *complex* (as opposed to "basic") interaction of all of its factors. We have a model of a tape head, a model of tape transport and a model of tape itself, including a bias oscillator and what not. These were created an calibrated for one "hypothetical" tape machine. But, as it turned out, its parameters could be adjusted to closely match the frequency response and non-linear behaviour of virtually any well serviced machine we threw at it. " (u-he Satin or Slate VTM? (Topic in the 'Effects' forum) | KVR Audio Forum)

Anyway, thanks for going to the trouble of writing all that up and posting it.
Old 19th April 2017
  #8
Lives for gear
 
🎧 5 years
I KNEW IT!

There was a reason I'm not happy with my emulators

Screw this BS I need tape!

GS help me out! What's the best 8 track I can buy while not also having to give HJ's behind the 7-11 to cover my mortgage? Is there anything worth while under $1000?
Old 19th April 2017
  #9
Gear Addict
 
PVJesper's Avatar
 
2 Reviews written
🎧 5 years
I have never done a recording on tape. Tape emulations are also not really my "colouring" of choice, mostly (maybe a track or two gets some on a mix). So i don't feel very personally invested in that regard.

But i thoroughly enjoyed reading your experiment and analysis. Thanks for taking the time to write it out!
Old 19th April 2017
  #10
Lives for gear
 
Robert Randolph's Avatar
 
🎧 15 years
Curious what your reference level was set at in Satin. It seems off compared to the others.

How exactly did you determine the 'recommended settings'? Just the manual?
Old 19th April 2017
  #11
Lives for gear
 
login's Avatar
 
2 Reviews written
🎧 5 years
Quote:
Originally Posted by Villberg View Post
I know there’s been a lot of blind testing comparisons between tape machines and their emulations. In my humble opinion, there’s a glaring problem with these kinds of comparisons - they rely entirely on the test subjects’ subjective perception of sound. There are many studies showing that the ears aren’t exactly precision instruments and there are a lot of variables ranging from the quality of the listening environment to the psychological states of the test subjects. I wanted to gather data for my study, not entirely unlike the analysis methods that the actual plugin developers use themselves to gather data for their algorithm needs. Since I do not develop plug-ins, I kind of had to go in blind and make it up as I went along, but I still think that I managed to find some meaningful data and even managed to achieve some conclusions that you might agree or disagree with here.
Congratulations on your research, I love when people comes with actual science in this industry full of smoke and mirrors.

The highlighted part is so true and still totally ignored by most folks.
Old 19th April 2017 | Show parent
  #12
Here for the gear
 
Quote:
Originally Posted by Robert Randolph View Post
Curious what your reference level was set at in Satin. It seems off compared to the others.

How exactly did you determine the 'recommended settings'? Just the manual?
I used a 1khz -18dBFS test tone and set the output so the test tone hit 0VU on all emulations and tape machines. There was virtually no tweaking needed for all the plugins, since all audio remained in the digital domain, and the plugins could be calibrated accordingly.

The recommended settings were determined by looking at the manual and presets. With Satin it was as simple as picking A827 15ips and dropping the 0VU ref to -18dbFS. With VTM I just picked the parameters that corresponded to the tape machines as closely as possible (Well, the noise slider was left to default -24db although I guess 0 would've been the "original" amount of noise).
Reelbus was a lot more trickier, since the preset "Swiss legends" and the info on the manual didn't line up. So I ended up choosing the preset and just putting the Device Adjustment controls to 100% instead of 0% (the scale is from 0-200%)
Old 20th April 2017
  #13
Gear Addict
 
🎧 15 years
@ Villberg Thank you for this.

I wonder how some Tape impulses through Acustica's Nebula (cdsm) measure up!

I would think they would would match actual tape better than the emulations.
Old 2nd May 2017 | Show parent
  #14
Here for the gear
 
OK, so my friend has an UAD card and some waves plug-ins and we spent a couple of hours last weekend repeating the tests on UAD Studer A800 & Ampex ATR-102 emulations, and with Waves J37 and Kramer master tape plugins. I've updated the zip-file in the OP to include these files for your comparison pleasure.

I will write about my findings later, but roughly inspecting the waveforms visually it seemed that both Waves and UAD featured transient response more similar to tape but also there was clearly more saturation going on. I did not yet calculate the crest factor values, but I will be sure to do that sometime soon.


Quote:
Originally Posted by wiz1der View Post
@ Villberg Thank you for this.

I wonder how some Tape impulses through Acustica's Nebula (cdsm) measure up!

I would think they would would match actual tape better than the emulations.
If you own Nebula, you could run through the test yourself and post the resulting .wav file back here so I can add it to the comparison.
Old 2nd May 2017 | Show parent
  #15
Gear Maniac
 
🎧 5 years
Quote:
Originally Posted by terrible.dee View Post
I KNEW IT!

There was a reason I'm not happy with my emulators

Screw this BS I need tape!

GS help me out! What's the best 8 track I can buy while not also having to give HJ's behind the 7-11 to cover my mortgage? Is there anything worth while under $1000?
I just bought an Otari MX5050 8-track half inch machine. I'm still trying things out, so far I'm impressed...
Old 3rd May 2017 | Show parent
  #16
Gear Maniac
 
🎧 5 years
Quote:
Originally Posted by Villberg View Post
you could run through the test yourself
I'll have a go, I'm downloading your zip atm...

Don't know when I'll have time to do it, but I have a Revox G36, a Revox A77HS, and an Otari MX5050 8-track to offer, and I also have an old Babyface, so it should make for an interesting comparison.

I have seen similar measurements in the past, but I can't help thinking there is some other value we could quantify that we haven't thought of yet...

Frequency response, phase response, dynamic response, transient response and... Parameter X...
Old 3rd May 2017
  #17
Lives for gear
 
Bouroki's Avatar
 
🎧 5 years
Reelbus highs are not weird, just faithful to the modeled Revox A77 mk4.

Quote:
Originally Posted by Villberg View Post
Ok so what causes real tape to handle the dynamic range differently? It’s certainly not compression. Nor is it limiting.
Circuit soft clipping? afaik a tape machine's circuitry has less headroom than the actual tape.

In Satin there is the headroom knob which is set to max by default, which totally puts the circuitry out of the equation - that should be brought down to a more realistic value. VTM probably models higher end machines with more headroom. In Reelbus with circuit clip at 100% it definitely exhibits noticeable transient softening as the input is driven more. So it makes sense that it is the one closest to the Otaris given your testing setup and I think if you increase the circuit clip beyond 100% it will get even closer.
Old 3rd May 2017
  #18
Urs
Lives for gear
 
🎧 15 years
Quote:
Originally Posted by Villberg View Post
Can this type of behavior even be modelled? I don’t know, that’s really a question for the really people behind the algorithms!
Indeed, and we would have loved to assist you in your presumptions before you posted this.

Unfortunately I can't read the whole story today but I was pointed to this thread, and this thread exists, so why don't you sum up your cirticism on Satin, send it to our support team and give us a chance to help you understand it better?
Old 3rd May 2017
  #19
qwe
Lives for gear
 
Quote:
Originally Posted by Villberg View Post
Can this type of behavior even be modelled? I don’t know, that’s really a question for the really people behind the algorithms!
http://uhedownloads.heckmannaudiogmb...user-guide.pdf

Sascha Eversmeier: "we're using a model consisting of virtual coils, hysteris loop, HF bias and eddy currents / self-erasure" (https://www.kvraudio.com/forum/viewtopic.php?p=5487254)


And once again... Paul Frindle on tape artifacts: https://www.gearslutz.com/board/12399465-post1162.html

"So first of all, you cannot record HF at as high a level as LF because the tape will saturate at lower levels. You can confirm this by trying to record white noise; to get playback to sound the same as the input you will need to record about -10dB below the operating level to achieve this.
"Furthermore loads of HF from the record head will push the tape into self erasure This means it's easy to saturate the HF and cause it to distort at HF and even cause the HF to self-erase as well. But this isn't ordinary distortion like saturating electronics - it has a 'soft' kind of 'splashy' effect often heard in over recorded cymbals etc."


Hopefully, a few clues as to what's going on. :-)
Old 3rd May 2017 | Show parent
  #20
Lives for gear
 
comfortablynick's Avatar
 
🎧 5 years
Quote:
Originally Posted by Urs View Post
Indeed, and we would have loved to assist you in your presumptions before you posted this.

Unfortunately I can't read the whole story today but I was pointed to this thread, and this thread exists, so why don't you sum up your cirticism on Satin, send it to our support team and give us a chance to help you understand it better?
What a nice offer! This is a cool thread; I'm looking forward to hearing from U-He especially, since they obviously took their modeling of the tape machines very seriously.

I'm sure it's tricky to get the plugins set on a comparable "default" setting, since they are all modeling something different and have different controls. Matching the headroom is critical, and I would think Satin would do quite a bit more compressing given the correct settings. That's always what it sounds like it's doing when I turn the Headroom knob down.
Old 3rd May 2017
  #21
Gear Addict
 
an_expresso's Avatar
 
🎧 5 years
Hate to go off topic but just wanted to chime in and say I love when Gearslutz has threads like this with intellectual and helpful posts as opposed to people bitching about nothing.
Old 4th May 2017
  #22
Here for the gear
 
Quote:
Originally Posted by Urs View Post
Indeed, and we would have loved to assist you in your presumptions before you posted this.

Unfortunately I can't read the whole story today but I was pointed to this thread, and this thread exists, so why don't you sum up your cirticism on Satin, send it to our support team and give us a chance to help you understand it better?
I first got to say that I am humbled by your presence in this thread! U-he is really one of the best plug-in developers at the moment IMO (though I own none personally, since I'm a completionist and cannot really afford the whole collection right now ). I was actually in contact with your support and asked some help to find the optimal settings before conducting my tests. Rob was helpful enough to point me in the right direction, though he couldn't squeeze much out of Sascha

As to summing up the criticism I’m still really trying to find out what’s happening here and posted the results here so we could find out more as a collective. BTW. Satin is still in my top 3 when it comes to how it sounds like subjectively, along with VTM (slight bias there on my part ) and the UAD ATR -emulation.

Quote:
Originally Posted by Bouroki View Post
Reelbus highs are not weird, just faithful to the modeled Revox A77 mk4.


Circuit soft clipping? afaik a tape machine's circuitry has less headroom than the actual tape.

In Satin there is the headroom knob which is set to max by default, which totally puts the circuitry out of the equation - that should be brought down to a more realistic value. VTM probably models higher end machines with more headroom. In Reelbus with circuit clip at 100% it definitely exhibits noticeable transient softening as the input is driven more. So it makes sense that it is the one closest to the Otaris given your testing setup and I think if you increase the circuit clip beyond 100% it will get even closer.
In my opinion, Reelbus was actually maybe the farthest away of the bunch with the settings I used and sounded like a more lo-fi machine. Tape machines really tried to achieve a flat frequency response and well-kept and maintained machines do really get close to that. This applies to consumer-grade tape machines as well, like the Otari MX-5050 and the Revox too (it is after all a Studer-design). I'm guessing the machine that TB used to create model might have not been in it's prime, since I read through the blog posts he wrote about creating the plug-in and the measurements he had matched the behavior of the plug-in.

Now generally when testing plug-ins comparatively it’s really important to have the plug-ins set up in a comparable way, so I’ve decided to also include the screenshots of every plug-in I measured. You can see from the VU meter on the screenshots where my -18dbfs test tone was driving the plug-in (I now see, that VTM was a slight bit less, can’t remember if I compensated for that or not ).


Quote:
Originally Posted by CaptainMarc View Post
I'll have a go, I'm downloading your zip atm...

Don't know when I'll have time to do it, but I have a Revox G36, a Revox A77HS, and an Otari MX5050 8-track to offer, and I also have an old Babyface, so it should make for an interesting comparison.

I have seen similar measurements in the past, but I can't help thinking there is some other value we could quantify that we haven't thought of yet...

Frequency response, phase response, dynamic response, transient response and... Parameter X...
This is really cool, I’m really happy if we can add some real tape machines to the test!
Old 4th May 2017
  #23
Lives for gear
 
EvgenyStudio's Avatar
 
🎧 5 years
Thank you for your time! Hope to see UAD verdict soon with test results.
Old 4th May 2017 | Show parent
  #24
Lives for gear
 
Bouroki's Avatar
 
🎧 5 years
Quote:
Originally Posted by Villberg View Post
In my opinion, Reelbus was actually maybe the farthest away of the bunch with the settings I used and sounded like a more lo-fi machine. Tape machines really tried to achieve a flat frequency response and well-kept and maintained machines do really get close to that. This applies to consumer-grade tape machines as well, like the Otari MX-5050 and the Revox too (it is after all a Studer-design). I'm guessing the machine that TB used to create model might have not been in it's prime, since I read through the blog posts he wrote about creating the plug-in and the measurements he had matched the behavior of the plug-in.

Now generally when testing plug-ins comparatively it’s really important to have the plug-ins set up in a comparable way, so I’ve decided to also include the screenshots of every plug-in I measured. You can see from the VU meter on the screenshots where my -18dbfs test tone was driving the plug-in (I now see, that VTM was a slight bit less, can’t remember if I compensated for that or not ).
I was mainly talking about the crest factor analysis and the transient softening that was going on. Now that I've seen the screenshots it confirms that Satin & Reelbus were setup with extremely generous unrealistic headroom. "Circuit Clip" in Reelbus should be set to 100% for it to be exactly matching the specs of the machine that was modeled, while the "Headroom" knob in Satin should be brought down to at least the halfway point (I don't have Satin anymore so I can't remember exactly the headroom values, but you get the idea).
Old 4th May 2017
  #25
Lives for gear
 
🎧 10 years
Extremely informative and insightful. Thank you for you work on this. Did you develop a favorite over the course of your testing?

Quote:
Originally Posted by Villberg View Post
Hi! I’m a very long time lurker of this community but never really bothered to register, since I didn’t feel like I had much to add to the varied discussion in this forum. Now I maybe do have something to add, so hello everybody! A couple of months back I got my Bachelor’s degree in Music Technology. I wrote a thesis on a subject that’s been quite controversial here: how close are tape emulations to the real deal? The thesis is in my native language (Finnish) so I won’t even bother linking it here. However I thought it might be interesting for you guys if I more or less briefly went over the results of my study in this thread.

I’m very interested in this topic myself, since I’m a plugin hoarder and a strong advocate for doing everything ITB. Even my guitar setup live and in the studio is fully digital and modelling. With this in mind, I thought it was time to try and take a deeper, more critical look at some tape emulations to maybe broaden my worldview a little bit. That’s why I pitted my trusty emulations (and one that I’m interested in buying) against some well-kept tape machines I could get my hands on. The participants in the study were: Slate Digital Virtual Tape Machines, Toneboosters ReelBus, U-he Satin with tape machines: Otari MTR-12 and Otari MX-5050Bii.

The Study

I know there’s been a lot of blind testing comparisons between tape machines and their emulations. In my humble opinion, there’s a glaring problem with these kinds of comparisons - they rely entirely on the test subjects’ subjective perception of sound. There are many studies showing that the ears aren’t exactly precision instruments and there are a lot of variables ranging from the quality of the listening environment to the psychological states of the test subjects. I wanted to gather data for my study, not entirely unlike the analysis methods that the actual plugin developers use themselves to gather data for their algorithm needs. Since I do not develop plug-ins, I kind of had to go in blind and make it up as I went along, but I still think that I managed to find some meaningful data and even managed to achieve some conclusions that you might agree or disagree with here.

I used the program Fuzzmeasure to obtain graphs of frequency response, phase response, and THD at different input levels. Surprisingly, different input levels didn’t have much effect on the graphs, so 0dbVU impulses are what I’m going to show you here. Additionally, I prepared a bunch of different audio tracks in Pro Tools and sent them through the processors with, again, varied input levels (-3dBVU, 0dBVU, +2dBVU, +4dBVU and +5dBVU). In the test I included isolated bass drum and snare drum hits, an isolated bass track and a full spectrum mono mix with heavy transients. (Crosstalk was a territory I didn’t want to get into here, so everything was in mono).

For settings on the plug-ins I used the settings recommended by each plug-in developer. This included reading carefully through the manuals of each plug-in and finding the settings (and in some cases presets) that - in the developers view - corresponded with a well maintained 1/4’’ 2-track tape machine, 15ips speed, a modern tape model with plenty of headroom (such as the SM900) and a bias of the most neutral nature in terms of the frequency response. Both of the Otaris also hit these parameters. Of course I’m well aware that all the emulations have modelled different tape machines with different specs from the tape machines I compared them with but I was hoping that by having multiple emulations and multiple tape machines I could find some kinds of trends between the two. I tried to really ask around my area to find a real 2-track Studer for testing but even the pro studios that I contacted had sold theirs or had one that was fallen into a state of disrepair. I included the Pro Tools sessions at the bottom of this post so we could maybe expand on this together if there's interest for that sort of thing!

Some additional technical details: I used an RME Babyface (the original one) for gathering the data. With the tape machines I used the analog inputs and outputs compensating for the input’s frequency response, either by using Fuzzmeasure’s Automatic correction or Totalmix’s internal EQ. With the emulations, I used ADAT to connect digitally to another computer with an M-audio Profire 2626 interface (in slave mode) running the plug-ins in Logic (since Reelbus is AU only). The sample rate for everything was 44.1kHz since that still is what I, my colleagues and most of you guys use in your productions. All of the plug-ins oversample internally.

The Graphs



The Frequency Response

Comparing the frequency responses, I’d say they look pretty similar, except for Reelbus! (VTM’s graph is a bit lower because these are normalized curves and the exaggerated head bump is bringing the whole graph down) All the curves feature the head bump in the lower frequencies, a relatively flat midrange and a slight boost before the drop-off in the high end with the Reelbus acting strange on the high frequencies. As you can see, not even the two tape machines have their head bumps in the same frequencies (the MTR-12 has it’s head bump almost as low as VTM approaching 20Hz while the MX-5050 has a head bump near the 30Hz area.) With Satin you might lose some bass, and Reelbus really eats into the high end with the settings that I did the measurements. See, the best part with the emulations, is that their parameters are really adjustable: Want to tame the VTM’s ridiculous low end bump? No worries, just use the Bass Alignment control! Want to compensate for Reelbus’ weird high frequencies? Just use the Color Adjustment and Low-High emphasis controls. Want more bass with Satin? Increase the head bump. I bet I could’ve gotten the frequency plots resembling each other quite closely with a bit of tweaking. However, this was out of the scope of my study and I found it interesting to find out what the developers themselves consider as “authentic”.



The Phase Response

One reason tape sounds so nice is the way it “delays” low frequencies ie. turns their phase. This results in the warmth and roundness of the low end. This behavior is quite common in the analog domain since transformers and other components tend to do this sort of thing to the phase response. This behavior has also been successfully emulated in all of the plug-ins. The VTM’s graph looks a bit weird under the audible range, but I’m going to hazard a guess that the lowest frequencies get delayed so much that the graph can’t handle it. I don’t know if it’s really relevant or not. The Reelbus is again behaving weirdly on the high end. I guess this might be related to the weirdness of the frequency spectrum, too!



Harmonic Distortion

Ok, here’s where things get a little weird. The graphs vary wildly when it comes to harmonic distortion! The thicker line is the second harmonic, and the narrower line is the third harmonic. The two Otari’s resemble each other close enough but I really can’t say what’s going on with the emulations. However, I could verify that raising input level mostly affected the odd numbered harmonics in every tape machine and emulation, a trait that has been traditionally attributed to tape. This is the part of my study where I definitely felt like I was reaching a bit. I do not know if measuring Harmonic Distortion this way is even relevant to the sound of the tape or if this percentage type-graph was the best way to represent it. I put it here for completeness sake, maybe some of you might have some insight for it.

Test impulses and graphs are all fun and games, but isn’t the goal of these processors to make music? Let’s move on to the second part of my study. The Pro Tools sessions.

The Kicker

OK, let’s firstly get my subjective experience out of the way:

Listening in many different monitoring environments - subjectively - to my ears the difference between the two tape machines are of similar magnitude than the difference between one tape machine and one emulation. I could not reliably distinguish the real tape machines from the emulations while only using my ears!

Now we can try to focus on some quantifiable data by examining the waveforms closely. When I was examining the transient of the snare drum I came across this, see if you can spot it! The order for the tracks is: the original digital recording, Otari MX-5050Bii, Otari MTR-12, VTM, Satin, Reelbus:



The first transient of the snare drum is significantly “rounded” when sent through the actual tape. I tried to find this same type of transient rounding in the isolated electric bass guitar track, but could not find that sort of behavior: Since the transients were softer, the relation between the transient and the rest of the signal remained the same. Whenever there’s a fast overwhelming transient for example a drum hit, the tape tends to round this transient in a way the emulations do not!

I tried to think of a way to investigate this further when I remembered Bob Katz’s definition of crest factor (the difference between peak and RMS levels of the signal). I proceeded to analyze the dynamic range of the full spectrum mix that I had sent through the tape with different input levels: I took sample accurate selections of the clips of differing input levels and used the audio suite Gain plug-in to take the measurements of peak and RMS levels of said clips. I used the values to count the crest factor of my mix for each plug-in and tape machine across all the input levels that I used. Here’s the resulting table:



As you can see. The hotter you send the signal through the actual tape machines, the more they compress the dynamic range. The difference between -3dBVU and +5dBVU is a whopping 1,5dB in dynamic range! The difference with the emulations did not go over 0,3dB and most of that can even be attributed to the Gain plug-in’s rounding errors (see, even the original digital clip varies within 0,1dB). I know there might be a better way to analyze the signal’s transients relation to the rest of the signal but when it shows even by comparing 30 second clips with the crude Gain -plugin, I can only guess that the differences would be more apparent with more advanced tools.

Ok so what causes real tape to handle the dynamic range differently? It’s certainly not compression. Nor is it limiting. It seems more related to the difference in level over the domain of time - almost like the tape can’t really keep up with the faster high level transients - like the travel from the bottom of the waveform to the top takes too long for the tape while “drawing” the waveform and it realizes it needs to start heading back down already. I wouldn’t really know how to explain it better and it might even be that I’m way off with my analysis, but that’s why I’m also interested in what you guys have to say of all this. Can this type of behavior even be modelled? I don’t know, that’s really a question for the really people behind the algorithms!

We also need to ask how important is this reduction in the dynamic range of an audio signal while reaching for the authentic sound of tape. I think it boils down to what kind of aesthetic you’re going for. If you’re going for a vintage flavored, distorted track you might actually get by very well with an emulation, since they’ve nailed the frequency response and add the same flavor of harmonics when driven hard. However, I personally think that transient control is very important in all sorts of rhythm based music and I think that this weird rounding of the transients really helps and guides the engineer to make pleasant sounding and weighty recordings that are traditionally associated with the sound of tape.

If anything, I hope this thread maybe raises non flammable discussion on the topic, maybe even some solutions how we could capture this kind of transient response digitally, either by improving the algorithms some way or maybe with some completely other tools (such as a certain type of saturation, etc.) Thanks for bearing with me this far. English is not my first language, and I hope it doesn’t show too much. I also realize this is one hell of a first post to make and that’s why I’m going to get a beer now, cheers!

EDIT: Here's the Pro Tools file
Here's the the link to the test session if you want to take a closer look and listen! The samples for the drum-part of the test were performed by the awesome Jari Salminen of Poets of the Fall!
Old 5th May 2017 | Show parent
  #26
Here for the gear
 
Quote:
Originally Posted by Bouroki View Post
I was mainly talking about the crest factor analysis and the transient softening that was going on. Now that I've seen the screenshots it confirms that Satin & Reelbus were setup with extremely generous unrealistic headroom. "Circuit Clip" in Reelbus should be set to 100% for it to be exactly matching the specs of the machine that was modeled, while the "Headroom" knob in Satin should be brought down to at least the halfway point (I don't have Satin anymore so I can't remember exactly the headroom values, but you get the idea).
With Reelbus I started testing with circuit clip at 100% and immediately thought something was wrong with the setup, since the signal was heavily saturated even with a 0VU signal. Then I inspected presets like "Swiss legends", but that had circuit clip at 0% so I had to come up with a bit of a "magic number" for it and ended up on 10%. I can repeat the test with all controls at 100% but I'm guessing that will end up in distortion galore.

With Satin you're also correct, and that is something I missed during my first round of testing. With the headroom knob full on, I basically had 9db more extra-headroom than what would be the "realistic" amount.

I'll try to find some time this Sunday to check these out. Thanks for the awesome feedback

Quote:
Originally Posted by drichard View Post
Extremely informative and insightful. Thank you for you work on this. Did you develop a favorite over the course of your testing?
Well, the Otari MTR-12 sounded pretty nice Seriously though, my favorite plug-ins soundwise subjectively are VTM, Satin and suprisingly the UAD Ampex. That's still based on my subjective view and really hasn't got much to do with this analysis.
Old 5th May 2017
  #27
Lives for gear
 
1 Review written
🎧 10 years
I'd be happy to run stuff through satin for you. But I'd suggest maybe switching to reaper since I don't own pt. I'm mostly a logic and live user.

It would also be interesting to know what you heard back from uhe so we can make these endeavors educational.
Old 5th May 2017
  #28
Gear Guru
 
4 Reviews written
🎧 10 years
Cool thread! Very interested in this since I have Reelbus and haven't bought/tried the others. I grew up with tape, and found Reelbus convincing. I also have SKNote's Roundtone which I haven't explored fully....
Old 8th May 2017
  #29
Lives for gear
 
jeronimo's Avatar
Excellent post! Thanks for that!
Old 9th May 2017
  #30
Lives for gear
 
JanZoo's Avatar
 
🎧 10 years
Swiss Legend preset uses that Swis 15 IPS tape model, which is consumer grade A77 tape machine...

I guess you should use some more "high end" models, Glue 1 or 2, or Soft & Warm, I would like to find out what could the closest settings on Reelbus be for mimicking A827 or Atr102 sound
📝 Reply

Similar Threads

Thread / Thread Starter Replies / Views Last Post
replies: 345 views: 123491
Avatar for tamiya
tamiya 28th November 2014
replies: 2568 views: 444307
Avatar for The Beatsmith
The Beatsmith 2 weeks ago
replies: 175 views: 45548
Avatar for Rolleum
Rolleum 14th May 2019
replies: 112 views: 32646
Avatar for Sky_Michaels
Sky_Michaels 24th September 2019
Topic:
Post Reply

Welcome to the Gearslutz Pro Audio Community!

Registration benefits include:
  • The ability to reply to and create new discussions
  • Access to members-only giveaways & competitions
  • Interact with VIP industry experts in our guest Q&As
  • Access to members-only sub forum discussions
  • Access to members-only Chat Room
  • Get INSTANT ACCESS to the world's best private pro audio Classifieds for only USD $20/year
  • Promote your eBay auctions and Reverb.com listings for free
  • Remove this message!
You need an account to post a reply. Create a username and password below and an account will be created and your post entered.


 
 
Slide to join now Processing…
🖨️ Show Printable Version
✉️ Email this Page
🔍 Search thread
🎙️ View mentioned gear
Forum Jump
Forum Jump