Dithering causing overs?

[twoohfour]

Searched and searched and cannot find any solutions to this issue! Would very much appreciate any assistance...


Recorded/mixed project at 48khz/24bit, into 32bit for making fades in Wave Editor, then performing SRC to 44.1khz in Sample manager (still good up to this point)... Then, when doing Bit Depth reduction in Sample Manager (using MBIT+ dither to 16bit) all my tracks all of a sudden have overages. Some of them as high as 0.9 !!! The clipping is enough to be audible, and when converted to MP3 is deadly.

How/why is this happening? I've never experienced this before and have never had a problem with this, and I can't seem to find anybody else experiencing this as well.

Could it be a Sample Manager issue?

[Bob Olhsson]

That's classic inter-sample peaking. If you drop the level a few tenths it'll go away and can't also overload D to A and lossy file encoding filters.

[twoohfour]

So, should I just put a -0.3 gain plugin into Sample Manager?

Won't that be generally bad thing to do to the audio?

[Bob Olhsson]

Originally Posted by twoohfour

So, should I just put a -0.3 gain plugin into Sample Manager?

Won't that be generally bad thing to do to the audio?

I'd put it into the limiter settings that create the files at the original sample rate.

[twoohfour]

Originally Posted by Bob Olhsson

I'd put it into the limiter settings that create the files at the original sample rate.

Oh wow, that means re-printing all the masters and writing all new fades... learn something new every day.... Always worked at 44.1 before this project...


EDIT: For anyone that reads this in the future, I found out it was the SRC causing the overs, *not the dither* ... apparently Sample Manager just wasn't updating the Max volume after performing sample rate conversion...

SRC before limiting seems to be the cure.

[Alexey Lukin]

Dither can add 0.01 dB to your peak levels, but not more than that. SRC definitely adds more.

[windmillsound]

Originally Posted by twoohfour

Oh wow, that means re-printing all the masters and writing all new fades... learn something new every day.... Always worked at 44.1 before this project...


EDIT: For anyone that reads this in the future, I found out it was the SRC causing the overs, *not the dither* ... apparently Sample Manager just wasn't updating the Max volume after performing sample rate conversion...

SRC before limiting seems to be the cure.

I'm pretty sure that Bob Katz mastering book says that inter sample overs amount to about 0.3dB - at any sample rate. So I never print a mix with my limiter at more than -0.3dB. If I'm ultimately going to mp3 I'll do it more like -0.6dB as I find higher levels cause audible distortion in the conversion.

[jamesp]

Originally Posted by windmillsound

I'm pretty sure that Bob Katz mastering book says that inter sample overs amount to about 0.3dB - at any sample rate.

Theoretically they can go much higher than that - possibly more than 6dB in some special cases although I'm sure Alexey knows more about this than I do.

James.

[Richard Black]

Theoretically, intersample overs can go infinitely high. On a track of an hour's duration, I'm pretty sure I remember calculating that they can exceed 20dB. But that makes a bunch of utterly unrealistic assumptions about waveforms and filter impulse response lengths. In practice, with regular music, I'd be surprised ever to see any evidence for more than the 0.3dB that Bob Katz mentions. In the classical music that I mostly record and work with, I don't recall even seeing 0.1dB, with one exception: if a waveform is clipped anyway you can get some pretty big overs, but in that case you might as well clip it the hell back down again 'cos it won't sound any different!

[Greg Reierson]

Originally Posted by Richard Black

I'd be surprised ever to see any evidence for more than the 0.3dB that Bob Katz mentions.

It can go much higher than that with severely clipped modern pop/rock sources. Anything done to create overs ITB raises hell with conversion back to analog. Of course, that has become part of the sound now...

I run my system at the native rate. If >44k1 I'll process at the native rate, limit to taste, SRC* and then run a safety limiter at the end to catch any potential overs and to makeup for gain losses in the SRC.

*The Voxengo R8brain Pro has a 'prevent overs' option that pads the result down enough to prevent it from going over.


GR

[Black Belt]

Alexey - So trying to get some facts straight...can both SRC and Dither potentially add gain? And in light of this, is it best to SRC>Limmit/Dither as the last part of the digital chain to prevent possible intersample-overs resulting from SRC/dither?

[Alexey Lukin]

SRC can add ~0.3 dB, dither can add ~0.001 dB. So, the best chain usually is:
1. SRC
2. Limiter
3. Dither.

[princeplanet]

Originally Posted by Alexey Lukin

SRC can add ~0.3 dB, dither can add ~0.001 dB. So, the best chain usually is:
1. SRC
2. Limiter
3. Dither.

Wouldn't limiting ITB sound better at a higher sample rate? Then perhaps lowering the result 0.3 dB before dithering?

[princeplanet]

Originally Posted by princeplanet

Wouldn't limiting ITB sound better at a higher sample rate? Then perhaps lowering the result 0.3 dB before dithering?

still waiting to see if someone has a strong opinion re the above.... anyone?

[Bob Olhsson]

Depends on the limiter design. The original Digidesign limiter in Sound Designer II up sampled the side-chain which worked well. I'd expect this to be SOP although there is always good reason to be skeptical of developers.

[twoohfour]

Many modern limiters have built in "Oversampling", so perhaps your thinking is right.

SRC being this much of a bummer though, it's hard to imagine that it's worth it to work at 96k... i've heard from some guys (guys with grammy's) that 44.1 is fine and dandy...

[Adam Dempsey]

Originally Posted by twoohfour

SRC being this much of a bummer though, it's hard to imagine that it's worth it to work at 96k... i've heard from some guys (guys with grammy's) that 44.1 is fine and dandy...

I'd say that's a bit of a misnomer... a great mix at 44.1k (even 16 bit) will always trump an average mix at 96k. But all else being equal, whatever the native sample rate will sound best, avoiding any extra sample rate conversion (and bearing in mind that 44.1k is only relevant for compact disc... and iTunes still – unfortunately – although they will now accept 24/native rate files from which they convert to 44.1k AAC "plus").

[jayfrigo]

Limiting and non-linear processing in general can often sound better at higher sample rates (for sound technical reasons, not only subjective), though I'm thinking more of 88.2 or 96 rather than 48. Still, I'd personally do the limiting with a slightly reduced output ceiling, followed by SRC and finally dithering. My typical chain includes outboard limiting (t.c. 6000 MD4) at 2X rate followed by a real-time synchronous hardware SRC into the DAW at 24 bits with dither last.

[Rossd25]

This may be slightly unrelated.

Sometimes Logic shows -.2 while Ozone's meters show -.3.

I set my ceiling at -.3 and I trust Ozone's meters over Logic. And there is the same .1 decibel of a difference for the commercial CD tracks that I have imported into Logic as well. I am wondering if anyone else is having this problem.