Originally Posted by j_j
The worst problem SHOULD BE keeping some 2dB or so of headroom to avoid clipping of the decoded audio. This holds for both MP3 and AAC, bearing in mind that some encoders enforce this by reducing input levels.
Hi JJ, I have learned and learn a lot from you, THANKS!
I think, in this case, the problem comes from the decoder and not the encoder.
For example, iTunes (the computer software), reconstruct audio clipping at 0 dBFS.
This is wrong. It would be much better to reconstruct to 32 float, normalize to 0 dBFS (maybe taking into account ISP's), and then going to 24 bit for the DA converter.
I have tested the new iTunes Mastering suite (except afclip because it does not work in my computer) and you can encode AAC files up to +12 dBFS without distortion (AAC from 32 bit float, file reconstructed with Wave Editor). A 1 kHz sine wave starts to distort at +14 dBFS (realtime @ 44.1 kHz).
So the problem, at least with the iTunes AAC encoder, is not on the input.
Clipping is fatal for lossy audio. But clipping is not 0 dBFS. Many masters clip at -0.1 dBFS or -0.3 dBFS. So the solution, in my opinion, is not in the final level of the master but in the absence of "square wave" clipping. Clipping is not bad for linear PCM but it's mortal for lossy audio, so I think it's necessary to do a different master for lossy.
If you go to 0dBFS without clipping you will have better sound than clipping at -1 dBFS and encoding to lossy.
The ideal scenario will be if all decoders, and lossy audio players, reconstruct to 32 float with ISP normalization to 0 dBFS.
I think it's time to update the iTunes software... IMHO, of course.
A side note from my iTunes Mastering suite testing.
afclip: it does not work in my computer. The wave picture in the manual looks very dangerous for the speakers. Maybe it needs a WARNING!: don't play this file.
afconvert: works OK.
Droplet: fast and easy. Great.
RoundTripAAC: absolutely fantastic!
It works very well as intended and it could be used also for creative purposes in a mixing environment:
1. Put the same soundfile in two tracks.
2. First track Dry, channel fader -1 dB.
3. Second track plugin order: Volume plugin #1--> RoundTrippAAC (with delay compensation) --> Volume plugin#2. Channel fader about -15 dB.
Volume #1: +30 dB (from normalized 0dBFS audio). (AAC input distorts).
RoundTrippAAC: test different encoder type and bitrate (each one needs its own delay compensation). The AAC-HE is really funny at very low bitrates...
Volume #2: -30 dB.
Name for this processing chain: Warmaac (AAC Spectral "Warmifier").
Introducing lossy audio as an effect...?
SRC is phase linear.
At 96 kHz I have discovered some non linearities. At +18 dBFS 3rd harmonic predominant... tape/console emulation prior to SRC at 96 kHz?