Mastering, Clipping, Compression, RMS and the whole shebang

[Wilkesin]

In the last year, after numerous level threads, compression threads, ghetto home mastering, recording, tests, shootouts, watching meters etc I felt like i had a great working modus operandi. Then...

Some people were bitching on the Hold Steady message board about the mastering on their new record 'Stay Positive'. I'm a fan, and have been listening to the album a lot. I can hear that it was a bit squished, but it didn't seem like a 'worst offender' in the bunch of 'loud records' out these days.

I've also been revisiting an old favorite, Shiner's 'Starless' in recent times. It was released if i recall correctly 'round about 2000. To me, an obviously less smashed recording, and a bit easier on the ears to listen to the whole way through. So....

I load up some tracks in the ol' DAW and shock....The Hold Steady peaks at -0.1 dBFS and has an average RMS of about -5.5. The Shiner track has an RMS of about -12, but clips all to shit! over 300 sample clips through one song. Still, way more pleasing to the ear.

A few weeks ago i saw a post on another forum where some guy was advocating doing away with limiters and just clipping the hell out of the master bus. I laughed at this (seemingly absurd) thought....

now? I'm thinking cover the fucking meters, stop looking at anything and going with 'how does it sound'....and suddenly i understand why these threads inundate every audio message board on earth.

[ledogboy]

The Hold Steady peaks at -0.1 dBFS and has an average RMS of about -5.5.

Good gravy

[farview]

It isn't the limiters that make the one CD sound better than the other. It's the extra 7db of headoom and dynamics that make it sound better. Using a limiter on the quieter CD would have sounded just fine.

Brick wall limiters weren't as good as they are now. If you want a real comparison, get a CD from '89 or '90. (not a re-master) You won't find any clipping and the rms level will be between -12dbfs and -18dbfs, depending on the style of music.

The other problem you will run into is translation. Your converters will sound different when they are clipped than mine will, or the guy down the block, etc...

[MoreSpaceEcho]

The Hold Steady peaks at -0.1 dBFS and has an average RMS of about -5.5.

this is yr problem right here. i'm not the hugest fan of clipping, but i'll take clipped at -12RMS over anything at -5.5RMS any day.

[dynomike]

Lots of great sounding records have tons of clipping. Load up "gimme fiction" by spoon (mastered by ten jensen, i think? at sterling anyway) Tons of clipping. TONS! Sometimes it sounds more natural than peak shaping limiting. Hopefully the mastering engineer will use their ears to determine which sounds better for the album, though obviously sometimes they don't.

[Cellotron]

When clients request very high average levels for their mastered tracks (all too often these days) then generally some sacrifice to the fidelity needs to be made in order to increase the average level beyond a certain "sensible" point.

The tools available to making things "louder" are:
rebalancing the spectrum - simply eq'ing and generally moving the mids and possibly upper mids forwards gives the perception of things being "louder"
compression - to achieve greater density - either in the analog or digital realm - with usually analog hardware performing best for this task ime
peak limiting - either in the analog or digital realm - with digital brickwall limiters generally working best ime for this task
clipping - i.e. overloading an input so that wave forms have their higher transients chopped off and "flat topped" - either at the input of the ADC or using a digital gain stage - with both methods getting good results ime.

With clipping you tend to retain snap and punch of transients but at the sacrifice of increased distortion (i.e. crackling and fitzing when you really push this hard). With many digital peak limiters you generally get less distortion than when you clip but at the sacrifice of making the transients softer (i.e. the kicks and snares disappearing). Usually material that is already distorted but has heavy percussive elements that you wish to retain (i.e. heavy rock) can deal with heavy clipping way better than something that has a more pristine source track (i.e. chamber music).

Anyway - this is NOT a one size fits all situation - which technique or combination of techniques that will leave the least amount of artifacts depends entirely on the nature of the source track, the desired end results, the available tools and how hard you push the limits of these tools.

As far as intersample peak overs - some playback systems with better DAC's can deal with these fine - but many consumer systems and many conversion codecs (i.e. mp3 encoders) will get more distorted results when delaing with material with a constant amount of these. Generally this is why most ME's will set output ceilings of tracks to a small point below -0dBFs (usually anywhere from -0.1dBFs to -0.5dBFs) - especially if they use clipping to achieve high average levels.

Best regards,
Steve Berson

[Cellotron]

It isn't the limiters that make the one CD sound better than the other. It's the extra 7db of headoom and dynamics that make it sound better.

Agreed!!!!

Best regards,
Steve Berson

[Wilkesin]

Load up "gimme fiction" by spoon (mastered by ten jensen, i think? at sterling anyway) Tons of clipping. TONS!

Weird, also another personal favorite. i guess the tl;dr version of my post is that my eyes have been opened to the fact clipping doesnt mean something is ruined. If it comes to that versus having a brickwall limiter change the sound/balance of your mix, then it's a perfectly viable option. Not that I ever intended to indicate that i feel like records have to be loud, of course.

[Andy Peters]

I load up some tracks in the ol' DAW and shock....The Hold Steady peaks at -0.1 dBFS and has an average RMS of about -5.5. The Shiner track has an RMS of about -12, but clips all to shit! over 300 sample clips through one song. Still, way more pleasing to the ear.

I wouldn't trust the DAW display to accurately show the waveform.

It would be nice if the typical display actually used a sinc() reconstruction filter when displaying waveform data (at least when zoomed in) but about the only one that does the right thing is Cool Edit.

Load up my Delta function waveform sample. A delta function has infinite amplitude at time zero and has zero value for all other times. (At time zero is infinitesmal length.) Since our sampling limits us both in amplitude and time, we can use a function that is 0 dBFS at the very first sample and all other samples are zero.

The following 'scope trace is the reality:



What does your DAW show?

-a

[Wilkesin]

I wasn't looking at waveforms, I was just using Voxengo's Span and Sonalksis' Free G metering plugs. Though, I'm not sure if they show clips when one sample goes over, or three consecutive samples go over (which is what I believe is the norm IIRC).

addendum: I obviously mistranslated or misrepresented something in my first post, 'cuz on second look the Hold Steady stuff doesnt hit -5 RMS (I think I accidentally looked at peaks), it's more like -9 to -10 RMS.

[joel hamilton]

I recorded, mixed, and played on most of "Starless."
I didnt use a ton of compression when mixing that record. We wanted it to breathe a bit, not because of any of the canned "loudness wars" statements that get made on the interweb like "headroom is the ONLY way to make something awesome. I hate compression. Mastering guys fuck everything up!!!" its like an angry mob, and I dont agree anyway, and the debate was not raging in 1999 when I was recording that stuff anyway.

I dont know the new hold steady record well yet, but times have changed, for sure. not for the better, not for the worse, just changed. Do you enjoy the hold steady record? if you do, then it is "working."
We tracked the Shiner record to two MCI JH24 2" machines, linked with a timeline Lynx sync system. We mixed to an MCI 1/2" analog deck. The whole album was brought into sound designer II through a digidesign 442 (yes, 442) interface.The 4 channel, 16 bit interface.
We also mixed that record pretty quickly. Josh's guitars were added after I left the band. Unfortunately, my CD book was stolen that had my original CDR of that record, with all the guitars ust me and Al... bummer.

[dynomike]

I dont know the new hold steady record well yet, but times have changed, for sure. not for the better, not for the worse, just changed. Do you enjoy the hold steady record? if you do, then it is "working."

Have you heard it? It sounds empirically terrible.

http://hypem.com/search/the%20hold%20steady/1/

[wedge]

How does one go about checking RMS levels in ProTools? Any freebee utilities and the like? I've a PC...

[Mark Alan Miller]

That delta function is interesting.
I wonder why, though, that "reality" on the 'scope' has distortion both following, and especially preceding the pulse?

Frankly, from your description of the delta function (obviously within the paramaters of 96k 16 bit data) the display on my editor was exactly what I expected. -96dbfs, then one sample at 0dbfs, then -96dbfs again. Obviously with a ramp up to and down from the 0dbfs point occurring exactly for the duration of two samples. That to me seems more like the 'reality' of a delta function within said parameters.
(Well, the sample was actually somewhere around/just below 0dbfs, as my peak meters read -0.1 and -0.2 db, L/R, respectively. I redrew the peaks up to zero to make sure I wasn't imagining things and got a true 0dbfs reading. Just an observation.)



On another note, a -5.5dbfs RMS level for a song is gonna almost always be destroyed dynamically, IMO. Unless you're only measuring, say, the loudest moments of the loudest section of the piece.
On the note of measuring RMS levels, remember that silences before and after the song, or long fadeouts, etc, will affect the RMS level across the duration of the song. A song with a really loud coda but really soft verses, might read quieter, for example, than a song that stays loud more or less consistently across its duration. But the coda of the first song might actually be far louder. And if you use RMS readings across entire durations of songs to help 'match' levels or gauge how much to compress/limit to attain some kind of "target level", these kinds of things will make the comparisons irrelevant.
One way to apples-and-apples it, kinda, is to only look at RMS levels, say, for the loudest sections (choruses?) for each song being compared. Just another thought...

[MoreSpaceEcho]

On another note, a -5.5dbfs RMS level for a song is gonna almost always be destroyed dynamically, IMO.

agreed. that kind of level is just stupid. even the most raging, brutal, steady state whatever-core type of music doesn't need to be that loud.