Harry wrote: ↑
June 20th, 2020, 22:03
What does this post-decode clipping exactly mean?
It means that after decoding a compressed audio file (primarily lossy formats) to lossless PCM your file might clip.
Not only does it depend on the decoder used, but also on the post FX that might be applied, e.g. limiting will essentially make sure you'll never see these clips. Resonic never uses post FX unless specifically enabled by the user, but generally tries to reproduce the audio as purely as possible as it's also intended for checking the quality of audio files.
What I understood technically so far is, that clipping in digital signal processing is (beside some effects) not good for signal quality, because every information about 0dB is cut off (clipped) and therefore lost. Because 0dB refers to the highest possible digital value, Right?
This is true for integer (PCM) audio, but floating-point audio can be scaled (mostly) losslessly like vector graphics, so it may go past 0dB. The whole signal chain in Resonic uses 32-bit floating-point processing.
Then what do positive values there exactly mean? When it's eg. 0.7 does that mean there was like "0.7dB" too much information on that signal?
It means that a peak was generated by the decoder that went past 0dB, i.e. it effectively overshot by 0.7dB.
What you need to know is that file formats like MP3 are lossy
because they lose a lot
of audio information during encoding. The decoder then essentially reproduces something that is as close to the original as possible, but it will always be an estimate at best with a lot of detail missing anyways. If a peak in a .wav file is just below 0dB it might just go over after encoding it to .mp3 and decoding it back to .wav (or playing it back, for that matter) because it was reconstructed from very little information.
In general experienced producers tend to normalize audio to something below 0dB before encoding to .mp3/.m4a to create a little bit of extra headroom for potential post-decode peaks. I typically go to -0.3dB as a rule of thumb, a bit lower for very loud stuff, as preparation for lossy versions. For lossless versions (e.g. FLAC) I leave it at -0.1 or -0.05 because post-decode clipping shouldn't happen there (all information is available in FLAC, the original signal will be fully reconstructed, hence it's a lossless
format). I've experienced MP3s in the past that went all the way up to +2.3dB after decoding. It's worth verifying MP3s with Resonic like that.
Continue reading here: https://www.gearslutz.com/board/masteri ... pping.html
Something similar you'd want to look up is "inter-sample peaks" and "true-peak metering". But keep in mind that this is NOT
related to mp3 post-decode clipping like some people might state.