Page 2 of 2

Re: An experiment in repairing clipped audio

Posted: Wed Oct 04, 2017 1:09 pm
by Chrispy_
Possibly, I usually use MPCHC as a player and that has "normalize" as an option, but also value in percent so it's a sliding scale rather than an "adjust gain to 100%" checkbox.

There's a seperate "regain volume" checkbox which I assumed was dynamic range compression compensation which compensated for the temporal reduction in volume caused by sudden bursts of high volume.

I don't really know what I'm talking about as I never actually master stuff myself, but it would be a whole lot better to my ears if the mastering was done so that clipping of the waveform never happened (using machine logic rather than error-prone human decision making) and that the range compression prevented was equally applied tastefully so that the difference between the average volume and peak volume was a sensible amount defined by a standard amount, rather than completely random and usually not useful.
  • If there's no range compression you get mumbled quiet dialogue that can't be heard properly, and ear-splitting action scenes.
  • If there's too much range compression, everything sounds flat and things like whispering and talking aren't obvious without context.

Surely there's an algorithm that can be standardised, such that <range compression/normalisation> results in audio tracks with plenty of dynamic range yet the peak volume spikes are never more than 2x the average perceived volume. Nobody wants to be deafened, blow their speakers, wake the sleeping babies or irate neighbours next door - it's totally unnecessary ;)

Re: An experiment in repairing clipped audio

Posted: Wed Oct 04, 2017 1:11 pm
by morphine
Surely there's an algorithm that can be standardised

Stop using logic, seriously :) All that we discussed is and has always been technically feasible. But producers and the bands routinely demand that it be louder and louder. Thankfully those wars are now dying down thanks to Spotify and Youtube saying "screw it" and applying a normalization filter across the board.

Re: An experiment in repairing clipped audio

Posted: Wed Oct 04, 2017 1:51 pm
by just brew it!
Yeah movie soundtracks are a different matter; I was coming at it from the direction of music.

Obviously a movie can't keep the rustling of leaves in the wind, normal conversation, jet engines, and bomb explosions all at their proper relative levels.

Re: An experiment in repairing clipped audio

Posted: Fri Jul 27, 2018 9:06 am
by just brew it!
Thread necro!

So I'm having another go at the de-clipping. Noticed a cheap used copy of the CD on Amazon the other day, and decided $6 was a bargain to get the lossless version. And yeah, the slight tilt I noticed in some of the clipped areas appears to have been an MP3 compression artifact; on the CD rip, the clipped areas are pegged hard against 100%.

Re: An experiment in repairing clipped audio

Posted: Sat Jul 28, 2018 5:48 pm
by Waco
just brew it! wrote:
Obviously a movie can't keep the rustling of leaves in the wind, normal conversation, jet engines, and bomb explosions all at their proper relative levels.

God, I wish they would try. Range compression in movies really bothers me - I would love to be able to set my max output at 140 dB and have everything reasonably ramp up to that rate. Unfortunately, you can't dampen the excitement of the Michael Bays of the world that'll go "You mean we can make that robot stomp around at max volume? DO IT NOW!" or "Make the bass drop really loud, max volume. We need to make sure they see that shiny thing over there." :oops:

Imax theaters used to be really good about ensuring that quiet things were quiet, and loud things were loud. They had some of the best mastering I've ever heard - especially the stuff formatted for "Omnimax", the domed theaters.


Also - for that reconstruction effort - I had no idea Audacity could do that. I have a few tracks I'll be cleaning up in the next few days!

Re: An experiment in repairing clipped audio

Posted: Sun Jul 29, 2018 8:55 am
by just brew it!
Waco wrote:
Also - for that reconstruction effort - I had no idea Audacity could do that. I have a few tracks I'll be cleaning up in the next few days!

Don't get your hopes up too high. While I'm quite happy with the results on the Shelby Lynne album, IMO it is a special case -- sparse production, with vocals high in the mix. I think these factors have really worked in my favor.

I've also experimented a little with other albums that exhibit audible clipping, and when there's a lot going on sonically the repair generally doesn't work anywhere near as well. At least when there's more going on, the clipping also tends to be less bothersome (but I'd still like to get rid if it if I could).

I wonder if you could train some sort of AI to look at the damaged waveforms and extrapolate the clipped off parts...