On Tue, 21 Apr 2020, Will Godfrey wrote:
Just an aside here. I'm in contact with a number of people who who work in digital recording extensively. Their recommendation is that you should regard your peak level as at least -3dB due to the potential of inter-sample peaks. With the headroom available from even 16bit audio, you can go down to -10dB and still not have any noise issues. To really appreciate this check here:
Lots of stuff out there on this. I came across an article that proclaims: "This worst-case example occurs when the audio tone is 1/4 of the sample rate." That is for 48000SR a 12Khz peak. This seems unreasonable to me though the idea that at 12Khz the peak could be as much as +3dbfs makes more sense. It would seem to me that intersample peak posibility varies with frequency and that the worst case would be 1/2 SR where the peak could be infinite (in which case it would be rendered as 0). However, if everything above 20k is filtered, that worst case is removed.
It would still seem that the worst case would be at the highest usable frequency (20Khz for audio). However, for recorded material (a signal that was once acoustic) this is less of a problem as most natual sounding microphones have a frequency responce that starts to drop off before 20Khz (closer to 16Khz) and so the signal close to 20Khz is already attenuated. The root frequency of the highest used note seems to be well under 10khz (from: https://www.zytrax.com/tech/audio/audio.html ) and so the range above 10Khz is mostly harmonics and therefore at lower levels anyway. (except for the trend in female vocals to use an exciter to "candie" the voice and make them sound 12 years old) This would indicate that even setting highest peak to -3 would be fine. This may be why 1/4 SR is quoted as worst case, it may be that in real world music, this is the highest frequency that that still has enough energy to peak over lower frequency sounds (they were testing with pre-recorded CDs).
If one is using internal soft synths, obvously any fundamental frequency can be generated or any harmonic can be emphasized beyond normal for artistic effect. I would like to see a graph that shows maximum intersample peak against frequency or against % sample rate.
I am sure there are some who feel that using 96Khz SR is the solution. Or who can hear a difference between 48k and 96k for this reason. Or maybe just use a more reasonable peak level to begin with. Considering 20db peaks in the audio above 85dBspl and a quiet listening space of 40dBspl... the dynamic range seems to be a maximum of about 65dB... lots of room within the 96dB range in 16 bit audio. (note that a really quiet studio may have a noise floor as low as 30dBspl)
So the reality is that there would be no harm in setting peaks to even -20 except that "loud sounds better" and people want all their CDs, sound files etc. to sound about the same level. Another fun day in the audio world.
-- Len Ovens www.ovenwerks.net _______________________________________________ Linux-audio-user mailing list Linux-audio-user@xxxxxxxxxxxxxxxxxxxx https://lists.linuxaudio.org/listinfo/linux-audio-user
