Ableton Forum

Khazul wrote:...but if you are going for higher reduction, then also decrease the peak level a little more - say down to -0.3dBfs as hard limiting close to digital full scale and then applying mp3 encoidng can cause heavy distortion on some playback devices.

Inter-sample modulation distortion

laird wrote:http://www.neurofunk.co.uk/images//frequency_ranges.pdf

a useful guide, not the bible.

its usually easier to use a notch EQ to boost, locate a "bad" range, then cut that range, than it is to hear sweeping EQ cuts.

thanks for that link then.

First: Thanks alot, Tarekith, for sharing your experience!

pixelbox wrote:Shit hot article Tarekith! This should totally be a big, fat, STICKY!

So...how do you train your ear? I know you said it takes time and practice, and I do know that before I've thought that things were sounding good in my little room, but playing it on a different system made it sound not so good, but there have been times where I thought what I did sounded good, and other (more experienced people) noticed flaws in the mix or they even thought it sounded crappy.

How do you listen in order to train your ear to what is "right"? Just listen to whatever mix you think sounds good?

I think there are two answers to this question depending on for whome you do the mixing/mastering to begin with.

1. If you want to reach a "standard" sound that most people are used to (at a given moment in history, as it will change by time anyway) and thus satisfy other people foremost then you have to listen to their music on their systems alot. After that you listen to your music on their systems alot and try to make out the differences. When back at your own flat you listen to both, their and your music on your system and try to make out the differences again.

2. If you want to satisfy your own listening needs or create your own unique sound without giving sh*t what other people are used to or expect then do it the way that you like it (and hope that anyone else might like it as well)!

You would use 1 if you are mixing/mastering for other people mainly, a mixure of 1 and 2 if you mix/master your own music for other people mainly and 2 if you mix/master for yourself mainly.

Well said man, there's nothing wrong with going after your own "sound" instead of always trying to get your tunes to sound like artist x, y, z.

I wonder a bit about levels, metering and bit-depth. If every bit represents 6db level, how can I set a level to say -3db or normalize to -0.2db? Actually I did make up my own thought about it, but wonder how exactly this is to be put into words?

I honestly don't know enough about the math's involved. My guess is that since the operations are actually carried out via floating point math internally in the app, it's possible to have this accuracy before once again converting to a fixed bit storage medium (ala 16bit files).

Thanks a lot - great info.

R.

Nah, must be something different since it's not bound to bouncing but you can just change the volume slider and hear it happen. Thinking about it again it came to me that it's not like you switch 6db per bit, but each increase in bit-depth values by another ^2 which increases the overal possible range by 6 db because quantization noise is lower.

16 bit (96 db) covers a rangte of 65336 steps/voltage levels (-32668 to 32667 when thinking of a waveform) while 15 bit (90 db) is only covers 32668 (-16334 to 16333). That means that 16 bit offers twice as many steps as 15 bit and it seems highly unlikely to me that the resolution increase of 16 bit vs. 15 bit is only put into 6 db more footroom. That would mean that the resolution of the other upper 90 db of both bit-depths would have the same resolution and that resolution get better and better the lower the signal/volume.

I investigated a bit further into this and in fact it seems that for PCM audio and all bit-depths used the values are spread evenly across the whole dynamic-range.

That would calculate to:

16 bit: 0,001464844 db per value step
15 bit: 0,002746582 db per value step

With 16bit representing 96 db and 15 bit representing 90 db 16 bit would offer nearly twice the resolution of 15bit.

Whatever, this means that you can quite savely lower the main output volume slider as far as resolution/bit-depth is concerned. Say you lower the volume by 6 db, that would cut off 6 db from a 16 bit signal, but still the resolution of the other 90 db (down to the noise-floor) remains that of a 16 bit signal! So in fact you do not lose 1 bit, you only lose 4096 steps out of 65536 while at 15 bit you lose 32768 steps!

It's still a good advice to leave the master level alone and mix the individual tracks because of the underlying precision of the DAW (mostly 32 bit, going towards 64 bit). Also some wise old engineer made clear that digital sub-groups/sub-preamps can behave very much like analog ones in that they can distort. In the digital world that would mean that the underlying algorithms sums up several tracks before calculating the master and when running several high volume/hot signal tracks that sum can hit the ceiling even before they are reached down to the master. Anyway it's always better to lower a 32/64 bit floating point track signal than to lower a 16 bit undithered master signal.

As always feel free to correct me, especialy since I have not digged deep enough into the actual specifications of PCM.

Like I said, I don't do math. What you typed looks very scientific and therefore must be true. A few people called me out on the 16bit thing for the master, I will change that in the future. I the meantime, I still maintain it's a better practice to ignore the master fader and just get the channel faders adjusted right while mixing. If for no other reason than it means i don't have to think about math.

Did I mention I hate math? Ironic considering I do engineering at my day job, and mastering at home I know.

Hehe, sorry mate, didn't mean to do so much math, but sometimes you need to know some things for sure. And given the extreme difference of losing whole bits out of a few to losing only some steps out of many seemed to be something I'd rather know about before pulling my faders.

Also I am quite with you on the "leave the master level alone" thing as I just put to my last post when I edited it (before I saw your answer).

To be honest there was a stage in my life where I was all about figuring this all out like you did, and did so many times. After awhile I just got sick of always doing it, before you know it you spend all your time doing nothing but fact checking numbers. Probably a good thing, unless you're a musician trying to write songs!

Anyway, huge thanks for the facts, I appreciate the clarification in the face of my recent laziness!

Actually I'd love not to be forced to deal with these things and have someone else do it...

Wouldn't know about that!

Tarekith, what's your opinion of recording at 24/88.2 versus recording at 24/96? Is there truth to the notion that dithering down from 88.2 to 44.1 will somehow work better (because it's an even multiple) than dithering down from 96 to 44.1?

Just curious.

Also, is 88.2 (or 96) worth the CPU hit and disk space over 44.1? Can people actually hear the difference? I understand the bit depth gives you dynamic range headroom. But what does a high sample rate give you once it's above the frequency we can hear at?

wow, outstanding post Tarekith