Oversampling and anti-aliasing filters...

Bringing a thread over here from Gearslutz…

Scott

Hi Scott,

Would it be possible to post up both the newer beta and previous beta versions along with the original version .dlls in same beta folder, so others could compare the differences in sound of the oversampling?

Anyways, I’ll just re-iterate my personal opinions on the differences to get the ball rolling here with regards to sound differences in oversample mode:

Original release version:

PLUS:
Has a yummy way of glueing all the sounds together in a very thick way.

MINUS:
Totally different sound/dynamic response to same settings with oversampling off.
Heaviest high freq rolloff down 8-10db @20khz

First beta version posted

PLUS:
Less high freq rolloff than original release version.
Good all round sound, nice snappy punchy transients
Closer to sound of oversamp off

MINUS:
Still a slight drop of -3db @ 20khz

Latest beta (March 26) posted:

PLUS:
Almost no drop off in high freq response

MINUS:
Sound quality not as good as older beta or orig release version with oversamp ON
Stereo image seems to get constricted more, not pleasent :confused:

My personal choice if only one to be used: previous beta, definitely not latest beta.

In an ideal world: previous beta AND original oversampling mode available in single plug, offering two very different dynamically behaving options (given that the difference in dynamic behaviour is soo significant and clearly noticeable)

edit I did some audio comparison examples on my website of all 3 versions using extreme settings (ie not at all ideal, but highlighting the differences described)

musicprogressive.com/bomb/

Older = original release version. Beta = previous beta version. Betav2 = todays new beta version

I REALLY need you to define “lower quality”…I’m a little perturbed by your choice of words, because we’ve got BETTER frequency response, and still have little or no aliasing…how does that make quality lower? Saying “I like this one less than that one” is perfectly fine, everyone has opinions, but when you start throwing the “Q” word around, I would prefer that you back it up with the OTHER “Q” word…Quantitative (measurable) data.

I am researching the issue and will continue to post here with results.

Scott

Well one thing is clear, if you’re only altering the filter slopeness, it IS definitely changing dynamic behaviour, comparing the original release version oversampling to either of the new betas, one can clearly hear a massive difference in pump and snap. This is not some subtle ‘need bat ears’ difference.

So if the filter is the only thing being tweaked and not any other part of the code, its not just affecting high freq content or aliasing, its totally altering the plugins dynamic behaviour. The audio examples I posted sound totally different to each other (though granted the 2 beta versions behaviour is closer together than compared to the original more gluey version)

Comparing just the new beta to prev beta, its like the new beta just sounds a bit more constricted but not in a good way whereas the older beta had more sense of air about it (despite having the very slight high freq rolloff).

I don’t have a clue about coding, or whats going on under the hood, all I know is what I hear, on both speakers (Dynaudio BM6A’s in acoustically treated room) and headphones (Sennheiser HD600 into Earmax OTL headphone amp)

As test files are nowhere near nulling each other it definitely isn’t a placebo effect either but a very discernable difference in dynamic response. If I was to do a null test on older and newer beta versions via the inverse phase test, and all I could hear as a difference was very low level distortion and high frequencies then yes, i would be inclined to agree. But instead null testing actually shows a very different sounding dynamic response, backing up what I’m hearing.

So imho, the flat freq response isn’t worth the inferior dynamic response. Older beta sounded better than this newer one for oversampling mode.

Try this for me…place two 2nd-order (12 dB/octave) low-pass filters at 21 kHz (that will give you the equivalent of one 4th-order filter, which I was using in the original release) ahead of the latest beta of Bombardier. Does that recapture much of the sound of the original? It should be VERY close (depending on the implementation of the filters you use)…

I’m not doubting that the filters will have impact on dynamics…of course they will, ESPECIALLY in modes I, II, and III. They should have much less pronounced effect in modes IV and Flat. I’m just saying that I’m not changing any of the code in the dynamics engine…the only thing that is being changed is the filters that FEED that engine.

I’m not trying to argue with you…I’m just trying to get the best, most accurate results I can from the plugin.

Scott

Absolutely, no arguments at all here, if I didn’t love this plugin soo much I wouldn’t be spending all this time trying to ensure it sounds at its best just like you, we both have the same goal here :slight_smile:

Most of my tests have been in Punch mode I as like you say it highlights the differences the most.

But even inserting an eq before the plug doesn’t really get close to the behaviour. It just doesn’t sound the same.Also If I insert a 4th order filter to try and replicate the sound of the original release version, its not even close. In that first version the transient snap was soo wildly different using exactly the same settings, I have no idea whats causing that and I guess you’re not sure either, but totally TOTALLY different sound.

Again, your previous beta sounded much better to the new one, even using an eq to try and compensate (i used the EQuality EQ in digital, minimum, analog and linear phase modes).

I would still recommend going with the older beta, I really don’t think people will mind the very slight freq rolloff. Hell, a lot of people didn’t even notice the much heavier 4th order filter on the original release!

Besides, quite a few much loved and well known high quality analog compressors alter the freq response, often in the very top and bottom ends.

Okay, here’s another compromise attempt: Download from the same beta link again. This time, the original fourth-order filter is being applied to the detector input, but NOT to the actual audio signal (this is only in “Oversample” mode). This should almost directly replicate the “glue” dynamic behavior without affecting the frequency response of the output. At least one of my super-picky alpha testers tells me it’s working…

Scott

Damn you’re fast, I was just putting together an audio and image comparison of the 3 previous versions. lol.

Here it is anyways: Here you can visually see the difference between the orig 4th order and the 2 newer ones. Orig 4th order totally different.
The two beta versions ‘look’ almost identicle. But if you listen to the audio examples you hear the different dynamic behaviour.

One is inverse phase mixing new versus old beta (so 2nd order filter vrs almost flat), the other file I inverse phase mixed the original 4th order filter vrs almost flat.

Picture + 2 audio null tests here:

musicprogressive.com/bombcomp/

…and i’ll just go check that latest beta version you posted up now :slight_smile:

Ok, I am officially a dumbass, we’ll get to that in a sec.

The new compromise is on one hand pretty cool, its like a half way between the orig 4th order and latest flat response. Freq response stays flat, has some of the glue. The problem is that it sufficates the stereo image a bit when comparing to the no oversampled version, so not good.

Here’s the I’m a dumbass part.

When I was inserting an eq I was only trying to insert the eq BEFORE bombardier. Well, I just tried comparing the previous beta to the beta before that (the previous flat response vrs the 2nd order that I really liked). This time inserting the eq AFTER bombardier and… i could barely hear the difference, if at all.

It seems what was happening was that I was just preferring the slightly rolled off sound (and minimal phase smearing) of the 2nd order beta version, so essentially yesterdays 2nd order beta and todays previous flat response filter are the same).

In summary, I still prefer the way yesterdays beta with 2nd order filter and down -3db @ 20khz sounded, but using todays earlier beta + eq pretty much replicates that anyways and thus, is just as good.

So for purity I would probably go with the totally flat response beta you posted earlier today. Personally I’d stick with yesterdays 2nd older filter just cause I like the pleasing effect it has on the audio and its sense of stereo image (makes it sound bit wider than no oversampling, compared to this very latest detector input filter which seems to shrink it a little)

Hope that wasn’t too confusing, there’s soo many versions now its easy to get lost lol. I realise this has just been my own personal opinion and I certainly dont speak for all, but thank you for taking the time to listen and experiment!

I’d take latest version any day (I mean the one uploaded today, march 26th). I don’t hear any problem with the stereo image. Sound has more definition than original version (because of the lack of ‘roll off’) and it doesn’t sound as harsh/not glued as the previous 2 beta versions. I tried it on vocals, strings, drums, synths, etc., on busses and single tracks. Best sound I find among all other versions. Please keep latest version for next update, thanks.

Listen to the very latest version he uploaded on some complex stereo rich material and compare oversampling on to oversampling off. Don’t you find with oversampling on it constricts the stereo field just a smidgeon compared to no oversampling?. For me there’s bit more smoothness and sense of air with oversampling off on that new version.

I do agree though that it glues more.

here is a link to two relevant abx tests, which i have already shared with mercado (results were reported to scott).

cerberusaudio.com/b/Bomb-xBomb.zip

  1. vocal test (release) vs. (beta) , bombadier is on the vocal bus only.

  2. mastering test (release) vs. (beta), 1 instance of bombadier within a complex parallel chain.

  • one file in each set has an “x” in the file name; p.m. me to learn which is which.

Cool, downloading now, huge files!.

I take it this is the latest beta and not the previous 2 betas? (can get a bit confusing with all the difference versions).

Also did you included the non-oversampled version for comparison?. If not this would really help in comparing any subjective improvements over non-oversampled which is the entire point of having oversampling :slight_smile:

hmmm, mastering examples seem like they’ve been very heavily limited, or they’re just plain clipping, either way a more relevant test would be to hear a difference of only the behaviour of the bombardier in action as I’m guessing something else was clipping or limiting the signals, they’re pretty heavily squashed.

What was also surprising is that the difference between the files was really small, nulling down quite deep, so i’m guessing the bombardiers settings were pretty subtle. Add to that the sqaushing effect and it doesn’t really help assessing the differences properly, though I realise you were trying to represent a real-world use of the plug with those examples. More exaggerated tests reveal far more the differences though (such as the huge difference in transient snap between the orig release version and subsequent betas with oversampling on when pushed hard with fast attack time).

Also noticed same high freq response of two mastered versions. Orig release vers of bombardier had greater rolloff, was that compensated for, or is that just the other plugs in mastering chain smashing it up?.

Anyways, despite all that I’ll give my impressions on those files

For the mastering comparison, I prefer the BOMB version over the xBOMB version.

I found the xBOMB version to be a little more upfront, but the BOMB versions just sounded higher fidelity, more natural sense of depth and air.

Regarding the Vocal tracks.

Initially I was impressed more by the xBOMB version as the vocal cut through much more, that was until I realised that the vocal on the xBOMB version is almost 2db louder, which gives a false louder=better euphoric boost to it. So I can’t really judge the vocal ones because the vocal relative to the music volume is different, but ignoring its prescence over the mix I would prob lean towards the BOMB version again.

Differences were very subtle though. Null test on the masters nulled down by 50db!, main difference being a noise floor, perhaps one was dithered and the other one wasn’t?

Updated. 2.00.01 now being posted to website. Best compromise of all behaviors and across platforms is in place.

Mac builds will take another 5 minutes or so to finish copying to remote server, PC builds done. Dynamic behavior when oversampled should retain all the attributes of the “rolled off” version, while frequency response should show little or no actual roll off.

Scott

Scott, you absolutely nailed it. Best all round version, without the slightly suffocating effect that the previous beta had.

Awesome job. Thank you thank you thank you!! :mrgreen: