Once in a while a normalize a few lossless files which need some volume with Adobe Audtion or DBpoweramp. I want to achive the maximum volume without clipping of course. So can I be safe with setting a setting of 0db? Or should I stick with -1dB?
0dB should be fine. 0dbFS represents the digital maximum and there are no digital issues unless you try to go over that limit. Some people like to keep the peaks slightly below 0dB to allow for "inter-sample overs" (where the reconstructed signal goes over 0dB). It's not something I worry about, and I believe the DACs reconstruction filter should be able to handle these peaks without clipping, even though the DAC itself does not go over 0dB.
Once in a while a normalize a few lossless files which need some volume with Adobe Audtion or DBpoweramp. I want to achive the maximum volume without clipping of course. So can I be safe with setting a setting of 0db? Or should I stick with -1dB?
I've tested a goodly number of DACs, and its not uncommon for them to clip prematurly or start to distort before FS. Normalizing to 0.5 or 1 dB below FS doesn't hurt overall loudness, but it eliminated the possibility of these kinds of distoriton, that probably originate in the analog side of the DAC.
Thanks for the replies, Ill stick with -1db and the tracks that need normalization.
I've tested a goodly number of DACs, and its not uncommon for them to clip prematurly or start to distort before FS. Normalizing to 0.5 or 1 dB below FS doesn't hurt overall loudness, but it eliminated the possibility of these kinds of distoriton
I agree with this advice. I've never noticed clipping on CDs normalized to 0 dB. But I've heard from others, including Arny, that some CD players and DACs can clip and I believe them. Since the difference between 0 and -1 can't possibly matter, why risk it?
--Ethan
It's quite easy to document claims that CD players clip. Is there any evidence that can be presented?
From a thread on this very forum a few years ago, here's the link to t.c. electronics' PDF tech library:
http://www.tcelectronic.com/Default.asp?Id=12367#Mastering (http://www.tcelectronic.com/Default.asp?Id=12367#Mastering)
The "Mastering" section contains a couple papers that are particularly relevant: "0 dBFS+ Levels in Digital Mastering", which I referenced in that earlier thread, and another that I either hadn't noticed or was not on the website at the time, "Stop Counting Samples".
The latter paper is a very interesting one, as it addresses inter-sample over distortion not just in D/A stages, but also in sample-rate conversion and lossy codecs.
In regards to distortion in D/A's, I found this excerpt to be particularly interesting:
Many devices exhibit a prolonging effect every
time 0 dBFS+ is hit even briefly, thereby making a short
transient overload worse than it otherwise might have
been. The cause for the sustained distortion (often 150-
600 ms) is believed to be analog circuitry latch-up
and/or recursive filters.
It's quite easy to document claims that CD players clip. Is there any evidence that can be presented?
If you would have asked me about this about a decade back when I was testing equipment hot and heavy for my old PCAVTech web site, I could have drowned you with documentation. I suspect that there are far fewer DACs with this issue these days, or it is at least not as severe. IOW the premature clipping would be at FS - 0.1 dB as opposed to FS - 0.5 dB. That the problem has been common is reflected in the AES recommendations for testing computer audio interfaces - most testing is recommended to be done at FS -3 dB. Knowing how delta-sigma DACs work, it is almost a surprise that it happens.
Winamp used to clip with 0dB Ogg Vorbis files when dithering was enabled. Don't know if it's fixed now.
Two possible solutions:
A:
My advise is to not normalize at all but instead use something like ReplayGain and add playback volume tags.
B:
If that is not an option then normalize to 0dBFS encode to your preferred lossy format,
and then decode it by either using foobar2000 and the replaygain scanner there as it lists the track peaks before you apply the tags,
or decode into something like Adobe Audition (as 32bit float) and look at the peaks.
If the decoded lossy is either 0.99989 or 1.0 then you should be fine, if it is 1.001087 (anything above 1.0) then you need to normalize to something lower.
Sometimes -0.1dBFS is enough, and sometimes you need to do -1dBFS or lower.
Any modern DAC should have no issues with peaks at 0dBFS (aka 1.0 or -1.0 floating point), it's only when lossy formats are decoded and due to their (surprise) lossy nature that the peaks may go beyond full digital signal.
This will either cause clipping (usually inaudible *knock on wood*) and passed clipped to the DAC or they are passed as float (Vista and Windows 7 passes the audio as 32bit floats for example) and depending on the drivers and hardware may be passed all the way though to the DAC with peaks above 1.0
Heck even crappy old "pocket" cd players should handle 0dBFS peaks, it's just that many times the peak may actually extend beyond as the sample point may be slightly off center of the actual peak "point".
Any modern DAC should have no issues with peaks at 0dBFS (aka 1.0 or -1.0 floating point), it's only when lossy formats are decoded and due to their (surprise) lossy nature that the peaks may go beyond full digital signal.
If this isn't audible, why does it matter?
Any modern DAC should have no issues with peaks at 0dBFS (aka 1.0 or -1.0 floating point), it's only when lossy formats are decoded and due to their (surprise) lossy nature that the peaks may go beyond full digital signal.
If this isn't audible, why does it matter?
Because the OP seemed to want to normalize without distorting the audio, and I thought it would be nice if he knew that even if it isn't 0dBFS that it may still clip.
As I point out in the last sentence, even lossless might clip depending on the sample point.
I fail to see why any of this is an issue.
If it doesn't matter, why did you bother to post what reads to me as a slightly rude remark?
And I never said it wasn't audible, and I never said it was audible. As per HA rules only an ABX can actually back up such statement.
I took exception with your use of the word need in "If the decoded lossy is either 0.99989 or 1.0 then you should be fine, if it is 1.001087 (anything above 1.0) then you need to normalize to something lower." I'm sorry that I didn't include that in my quote earlier.
Seeing that the OP is talking about lossless, I'm wondering why you brought it up at all.
I took exception with your use of the word need in "If the decoded lossy is either 0.99989 or 1.0 then you should be fine, if it is 1.001087 (anything above 1.0) then you need to normalize to something lower." I'm sorry that I didn't include that in my quote earlier.
Seeing that the OP is talking about lossless, I'm wondering why you brought it up at all.
Point taken, maybe I should have dropped solution B as solution A is actually what my primary advise is in either case for lossy or lossless.
EDIT: Or maybe just written "might need" instead of "need" *shrug* (I know HA is picky on stuff, but really...*laughs*)
Your point taken as well. Provided that we put all this into perspective regarding audibility, you're right, it isn't a big deal.