1

I just noticed a couple of files I converted from FLAC to M4A (AAC) with Selene Media Encoder + Nero AAC now have substantial clipping and distortion in them even though the original FLAC had none and in fact didn't have a single sample touching 0 dBFS anywhere.

Is it normal for the pure and simple conversion to a lossy format to produce higher peaks (and in the case of normalized originals a real risk of clipping) or should I suspect problems with how the conversion is being performed by the particular software I'm using?

  • You could be seeing the reason that people don't master to 0dB, rounding errors & overshoot. It's just safer to master to -0.3 & not have to worry about it later. – Tetsujin Nov 1 '16 at 20:12
  • How close to 0dB does the FLAC get? Little increase in peak level are expected (that's why most people master to -0.1 or -0.3dBFS before lossy compression), but this sounds unusual. Are you sure there isn't a gain stage you're ignoring somewhere in the encoding process? – Linuxios Nov 2 '16 at 2:59
  • @Linuxios Most of the tracks peak above -0.1 dBFS. As I re-encoded about 3 albums just for this reason I noticed that only starting from about a -0.5 dB original gives me no-clip re-encodings in 99% of cases and for one song an exceptionally high peak was generated that clipped even when starting from a -1 dB original. I'm not sure there is or isn't a gain stage somewhere, as I don't know these programs, I just got them off the Internet, I didn't write them myself. That's why I started out by asking if the pure AAC encoding algorithm itself should be expected to produce this kind of change. – Don Joe Nov 2 '16 at 8:59
2

Yes, certain compression formats can result in higher peak levels than the same material pre-compression. -1dBTP ("true peak" level, relative to full scale) seems to be the accepted maximum safe level for pre-compressed material when considering this. It's important to ensure you are infact analysing true peak (aka. inter-sample peak) level and not sample peak level in this measurement, as adjacent samples can occur either side of an analog waveform peak, particulary in higher frequencies where wavelength is shorter. Even without compression, innacuracies with sample peak measurements can translate into issues when your digital material is converted back into analog for playback, which may not even manifest audibly until you play the material through low quality speakers.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.