Normalization before S/W Compression?
Home > Home Recording Forum > Recording Techniques > Normalization before S/W Compression?
Posted on Dec 01, 2003 02:43 pm
Bob
Member Since: Jun 20, 2003
I've been using the Sonic Timeworks Compressor X plugin in Sonar 2.2XL on most of my tracks and I'm generally pretty satisfied at how the sound is improved. Most of the time I have just used the presets. One thing I'm trying to understand better is a lot of these presets have fairly high threshold settings, eg their Vocal preset has a threshold of -7 dB. Generally when I record I try to allow myself 3-6 dB of head room, with fewer than say 5% of the peaks getting up to -3dB. So with the Vox compressor preset, it seems like most of the time I would only be getting about 1 dB of compression. My question is what levels are assumed to be in the recorded track for these presets to work? Should I normalize first?
[ Back to Top ]
Dec 01, 2003 03:25 pm Don't use the preset threshold levels. They have nothing to do with what YOU are compressing. Just set the threshold so it sounds good to you. Whoever made the preset has no idea how loud your signal is going to be, so they basically just pick a random threshold setting, expecting that you'll adjust it for whatever you're doing.
MinkusMazBane of All ExistenceMember
Since: Mar 27, 2003
Dec 01, 2003 04:00 pm i don't know how that plugin works, but it should also tell you exactly how much gain reduction is going on.
don't normalize. just get a strong handle on the part the compressor plays with vocals as opposed to a snare drum or whatever. and it's seriously all about ears. presets are interesting, but they are just blind suggestions.
Noize2uCzar of MidiAdministrator
Since: Apr 04, 2002
Dec 01, 2003 06:05 pm Yep agree with Minkus here. I dont normalize untill the very end.
WaltChief Cook and Bottle WasherMember
Since: May 10, 2002
Dec 01, 2003 11:16 pm Agree. Normalizing is a process specifically ment to bring the average level of a mix or set of mixes to a "printable" level. Now, I have cheated and used the normalizing function to amplify a low signal. Set the ceiling of the normalizer at say neg 6db to allow play room and it will amplify the signal evenly. Of course the signal to noise ratio will not change; i.e. the noise level will also amplify.
Noize2uCzar of MidiAdministrator
Since: Apr 04, 2002
Dec 02, 2003 08:56 am Indeed Walt, I guess I have cheated as well using the normalize to bring up a week signal. Usually on imported tracks or something of that nature. But as well as you stated it will indeed bring up the noise level as well if the track contains excess noise.
BobMember
Since: Jun 20, 2003
Dec 03, 2003 11:57 am OK guys, thanks again for the advice. Still learning all the in's & out's of mixing and level setting.
So here's a related SONAR question:
When I get several tracks recorded (say 12), and assuming I've left a little headroom on each track, eg the peaks are at say -4 dB, if I leave all the faders at 0 dB and the output fader at 0 dB I will most likely get some clipping in the output mix. What is the best way to avoid this:
(1) Leave the output level at 0dB and reduce all of the volume faders on the individual tracks to lower levels.
(2) Reduce the output main fader to a lower level, allowing the individual track faders to be set higher.
(3) Record the tracks at lower levels to begin with.
I'm busy with fader automation (using enevlopes) in the mix-down for the project I'm currently working-on, and I've used approach 1, above (output fader set to 0 dB), and I'm finding that with lots of tracks, the individual volume envelopes wind-up needing to be set pretty low. So now I'm thinking I should have used approach 2 (drop the output level & use higher envelopes). Any guidance on this would be appreciated.
Noize2uCzar of MidiAdministrator
Since: Apr 04, 2002
Dec 04, 2003 06:05 pm Well to tell the truth, if you follow the rules applied to analog equipment, yes it would be option 2 you would use. In most signal path situations, you reduce gain as you go down the signal chain. However, this does not really hold true in the digital realm. But that said, I do follow that rule of thumg even in the PC realm. I find the best situation for my tracks allowing some headroom, then if need be I will reduce the output signal of the master fader untill I get an acceptable level of output.
As far as recording the tracks at lower levels, there is a catch 22 there. You want to make sure and get them in as hot as possible, but at the same time, you dont want to take any chances of hitting the high end even for a fraction of a second. This will induce digital distortion, and that is not a very good thing. What I do is try and find a happy medium, I will try and keep the tracks about 3 to 6dB below clipping. And if I get the track in to hot, then reduce the input and try again untill it is good.
Hope that helps, but it sounds like you knew what you might hear anyway.
BobMember
Since: Jun 20, 2003
Dec 04, 2003 10:08 pm Thanks Noize. Sounds like a good answer. One of the other things I was concerned about, with running my individual recorded tracks hot (but not clipped) into the Sonar mixer and then bringing the mix down out of saturation using the output fader, was would there be some intermediate point in the Sonar signal flow that would overflow? I am concluding that the intermediate mixing is done with more than 24 bits of dynamic range so that will not be a problem. Have I got that right or am I out of my tree?
WaltChief Cook and Bottle WasherMember
Since: May 10, 2002
Dec 04, 2003 10:37 pm Wow Bob, You have some real good questios there. I hope Noise, Jues, Db, or etc. will notice and shed some light on them. I really don't know how your software app. handles volume samples above sample depth latatude. Wether all signals at max and above are registered as max ( creating loss of dynamics) or not registered causing cliping.
To be honest, once I start mix down, I no longer use my output meters in software. I trust my ears. I keep my individual channels out of the red and work on balance, stereo field, volume consistancy, tone, etc. Once mixed, I use a mastering program to bring the whole thing to the top for printing.
One comment on option three from two posts above recording at lower levels. Remember it's all about signal to noise ratio at the point the signal hits the A to D converters. It is not about saturation. If you are introducing noise by busting the signal, back off. Unlike old analog systems where you had to get extra signal to saturate tape, get above tape noise, anticipate, componant noise during mixdown, once into a sampled digital format, the cast is set. It is all about simple math making signals louder, softer, etc. The tinyest signal can be amplified to sound great if the noise floor is far enough below it.
Noize2uCzar of MidiAdministrator
Since: Apr 04, 2002
Dec 04, 2003 11:36 pm Good going Walt, you are so wise. As Walt said, in the digital domain there is no saturation, only clipping. And yes, Sonar as well as all the others go into digital clipping whe the input signal is to high, and that will cause noise. Sometimes it is not noticable untill the final mix comes in and you cant maintain a good level.
OK, now onto the overflow thing. Yes there is a side effect to the digital domain when dealing with multiple tracks. When you have maybe 2 to 4 tracks things dont stack up to badly and you can generally mix down without to much ado. But when you get more then that, what happens is the signals now all combine and become audibly as well as digitally louder. Hence the reduction of the output level to allow for a better set of stereo mixdown tracks. It may take some experimenting the first few times but you will find that happy medium between the track levels and the final output levels. I used to get frustrated by it as each piece of music is a bit differant in the way the tracks stack up, and how you mix the levels to get the correct output level. But now I just take it for granted and give myself just that little bit of extra headroom both in and outbound.
If I may ask, are you exporting directly out of Sonar as a wav file? And if so, are you using any other software to pre-master or master your final 2 track mix? If you arent using any other software editer for the final mix, you can mix the tracks down right into another set of tracks in the same project, then Mute all the other tracks while you work on the final sound of the 2 track mix, then export that finished product out to wav and burn away.
And as Walt said, you are asking some great questions here, keep em coming.
Peace
Noize 2 U
WaltChief Cook and Bottle WasherMember
Since: May 10, 2002
Dec 07, 2003 10:57 am Noise,
Sorry to bug and be anoying and well, that's a lie, I love it actually. (sinister snicker)
Here's the piece I would like defined. I have my 10 tracks in Sonar, Cubase, EIEIO#3, etc. I get them all balanced out and I bring up my output meters and lo and behold their off the scale. I can think of two ways software could deal with this.
1. Assign those samples that are over the top no volume code or some unassigned code and effectively give them the actual sound of cliping.
2. Assign those too loud signals a max volume code effectively removing dymanic change from max volume and beyond.
What do the software algorithms do with over the top samples generated during mix down? I know as you mentioned that samples out of the A/D converters get assigned the nothing or gobeldy goop level code if they are too high causing audable cliping. I am assuming that this is because the software simply accepts the output of the A/D converters as it samples with no conditional statements involved.
BobMember
Since: Jun 20, 2003
Dec 07, 2003 01:46 pm Yeah Walt, that's sort of what I was trying to get a better handle on. I'm sure there's some literature out there that explains how these audio apps do their internal math so it's kind of silly for me to conjecture on how it works instead of doing my homework ... but I'll be silly anyway. I believe that when you record in 24 bit format and you max out an input level to last possible bit of available amplitude, that you are causing your AtoD to put out a code of +/- 2^23. This is represented as a fixed-point two's complement binary value. If you push the level the equivalent of 1 bit higher, the AtoD might either clip at the max value or wrap to a value of the opposite sign. Either way distortion results.
So lets say you stay just below +/- 2^23 on your individual tracks into Sonar but you start adding these together through internal mixing. The sum of these values is clearly going to exceed what can be represented in 24 bits. I've seen reference in a number of plugins that they do 64 bit processing, so they shouldn't get overloaded. I need to do some more looking at Sonar documentation to findout how many bits they use to represent on the main and aux busses. I gather that the output meters are showing you how the current mix fits within a value that can be represented in 24 bits, so that if you do a mixdown (bounce to track, etc) you can avoid clipping in the destination track (which can only hold a 24 bit value). When you export a 24 bit track to a 16-bit broadcast wav file then (Sonar) conversion S/W has to divide each sample by 2^8.
Some of this is probably obvious you folks that really know these recording apps inside & out, so please parden my rambling.
WaltChief Cook and Bottle WasherMember
Since: May 10, 2002
Dec 07, 2003 02:04 pm Yes Bob, I am being a bit lazy myself. Problem is with most manuals written in layman's terms only and also written as a marketing tool, that research could be a real pain. Suffice to say, I know how to deal with it at a functional level and my question is primarily accedemic. (If you hadn't noticed I pride myself in extreamly poor spelling). If I spot something in my reading I will post it for reference, but to be honest, I am not going to make the research a high priority at this point. Better for me to send the time practicing my basses. I have a session comming up where I will be playing both upright and guitar. I would really like to do a good job.
Noize2uCzar of MidiAdministrator
Since: Apr 04, 2002
Dec 07, 2003 07:12 pm Bob has been doing some homework indeed. The way each app processes the signal on the way out is just a bit differant with each app. but still very much the same. It is just the math that varies. But in essance the statement Bob made about the output meters is correct. If they indeed clip, then you will have distortion on the mixdown, either bouncing internally or exporting. And yes it does appear worse when exporting. But as he stated, when the tracks are balanced and the output is set to not excede upper levels, then all should be good without distortion.
That is partially why I asked about using 2 track editors or doing final master in Sonar. I will always give myself just that little bit of extra headroom when exporting to wav file only becuase I know I will max the file out once in the 2 track editor. I will allow a little less headroom when doing a quick 2 track mix in Sonar, only becuase I dont as a rule do my final 2 track in Sonar even though I very well could, I am just more acustomed to the tools I use in Wavelab and SoundForge.
WaltChief Cook and Bottle WasherMember
Since: May 10, 2002
Dec 07, 2003 10:48 pm Works for me, although I will admit the more mixing I do the closer to the top I get. I guess intuitively. I am noticing that once I get to the mastering app. and apply my 3db buffer on the top for playback compatability, I am just about maxed. Then it is just a decision on how much top end distortion I want to add to the piece to infer power or loud to the listener.
Really a shame in a sense. A clean mix will be interpreted as "amature" per the lack of added distortion in the mastering process. But by the same token it is all part of it. Interesting the psychology of listening.
WaltChief Cook and Bottle WasherMember
Since: May 10, 2002
Dec 08, 2003 12:14 am Bob,
I lied. I got to ruminating on the question and did a little reading. Without getting into codes and numbers here is how Cubase SX handles volume:
SX uses 32 bit float to it's fullest extent and offers virtually limiteless headspace. Therefore on each individule track a VU reading above 0 is actually not problematic at all.
The combined volume units as represented in the output section are also 32 bit float and are accurate within the digital relm with no distortion above 0 dbm.
Here's the kicker. The digital sent from the output section is subsiquently calibrated to the hardware e.g. D/A section. Outputs above 0 dbm will create clipping and distortion at the D/A section.
So in the digital domain no distortion occures regardless of volume units. However incoming distortion is recorded and retained as received from the A/D section and over driven signals will result in distortion on exit to the D/A section.
Makes perfect sense once I read it. Daaaaa. Oh well. Cubase recomends mix individual tracks as you will and adjust the output to accomidate for the D/A section, i.e. not above 0.
Now maybe I can go to bed. I hate ruminating on unanswered questions... maybe I need to increase my medication... Or is it those space people again? Yup time for bed!
Noize2uCzar of MidiAdministrator
Since: Apr 04, 2002
Dec 08, 2003 04:28 pm Now that you bring that up Walt, I believe Sonar 3 as well uses the 32 bit floating point. And again it does handle the clipping about the same way. And it all boils down to the end result coming out the D/A. The levels must be mantained below the clipping level on mix down.
WaltChief Cook and Bottle WasherMember
Since: May 10, 2002
Dec 08, 2003 07:50 pm Ya know, come to think of it. I'll bet that is how I was able to retain so much power in the live mix I did of Project Dead. I mixed it down very hot but 48K 32Bit. Improted that into a mastering montage and mastered to -3db capped while dithering to 44.1K 16bit. By maintaining the 32bit throughout, I was able to probably work in regions that would have been pure distortion in the past.... Interesting.
BobMember
Since: Jun 20, 2003
Dec 09, 2003 12:48 am OK great Walt, Noize,
So this brings me back around to my original question, and sorry to beat a dead horse, but If nothing downstream (in Sonar) is going to distort even when all of your recorded tracks are max-ed out (but not clipped, mind you), then what is the downside to normalizing your tracks? It seems like this would give you more predictability on how to mix and set up effects. Then in the end just gain-down the output appropriately. But I may be missing something here because I did that on my last project and I either wound up with distortion or for lack of a better term 'sonic overload'. The normalization may have bit me ... or ... it may have been just poor EQ'ing on my mix (to notch-out a hole for each instrument). I've seen blueninjastar's writeup on EQ. I'll give it a try.
WaltChief Cook and Bottle WasherMember
Since: May 10, 2002
Dec 09, 2003 06:48 am Well Bob,
The only downside would be an extra step for each track. If the extra step works for you then obviously; enjoy! The routine is simply linier amplification of the signal with the hottest spike set to 0dB. As far as your sonic overload (I'll choose this per no reason to believe any distortion is involved) yes, the technique of notching tha Blue deliniated is a good technique. So is using your stereo field if you are sure that the piece will not be played mono. Down the list of things that can cause problems are inverted phases and altered phases (combing).
Per my perception this is what makes the endeor fun. If it was a simple recipe say for microwave popcorn, I would do it once and be bored. I love the hobby/endevor because I can draw on so many things. To be good at this one can end up with a very good knowledge of sonics, electronics, computer science, wave transmission, pyhsological perception, et-infinum. Oh yea, and a lot of experience / practice.
Listen every step of the way. My bugaboo right now is the harmonic exciter for example. I have yet to successfully use one. I don't like the effect yet. To my ear it does nothing but distort. Eventually I may learn to use it effectively based on my jouney into recording to date.
Noize2uCzar of MidiAdministrator
Since: Apr 04, 2002
Dec 10, 2003 10:29 pm I see where your going with the maxxed out tracks thing and the normalizing on each track. But if each track is normalized to its max level, then it is possible there is no mix left when you mix it down. I may be reaching here, but when I do a mix, very few of my tracks are at the same level, fader wise on my software mixer. But I can still, if the tracks are hot enough end up with overload on the output.
And yes, EQ could play a large part in the sound of the final mix as well. It sounds though as you might have a good path to follow. Check out Blueninjas article and see what it brings.
