From: Wesley C. <we...@se...> - 2009-08-29 04:10:03
|
I've got some recordings of concerts to which I'd like to apply range compression. I've found that if I just increase the volume on the recording, drums and clapping tend to blow out the ear drums while listening with headphones... until I found sox's handy compand command! (I'd never been able to get Audacity's range compression to work, but that's another problem) I'm using this filter from the sox man page "to make a piece of music with both quiet and loud passages suitable for listening to in a noisy environment such as a moving vehicle": compand 0.3,1 6:-70,-60,-20 -5 -90 0.2 And it seems to work reasonably well. Not being an audio engineer myself, I don't really understand what any of those numbers mean. Can someone suggest a better set of values to try? Thanks, Wes |
From: Fmiser <fm...@gm...> - 2009-08-29 06:17:03
|
> Wesley Chow wrote: > compand 0.3,1 6:-70,-60,-20 -5 -90 0.2 > > And it seems to work reasonably well. Not being an audio > engineer myself, I don't really understand what any of those > numbers mean. Can someone suggest a better set of values to > try? It is highly subjective - so "better" is tricky to achieve. What is a good set of values can vary depending on what you like to hear, did it come from cassette tape or CD, type of music, instruments (snare drum solo will be different than a pan flute solo), etc. But I'll try to explain in non-engineer terms what those numbers mean. Hopefully then you can do some educated experimenting to figure out what works best for you. *smiles* * The first two (0.3,1) are "attack" and "release". The compander is effectively a robot controlling a volume control. These values determine how fast the robot is allowed to turn the knob. "Attack" is how fast it's allowed to be turned down. "Release" is how fast it can be turned back up. If the attack is set to fast (small number), then every sharp, snap-like sound will be "caught". That's good if the snare drum hits are what need to be control. But not so good if you just want the soft passages louder. If the release is set to short (small number), it's more likely there will be "pumping" - that is, it's audible the volume is being turned down and up. If your goal is to make the quiet passages louder, and keep the clapping _sections_ from overwhelming - try large-ish values. Experiment - but you probably want them to stay near each other. "0.02,5" will probably not be what you want. * The next one is "knee" This is how "soft" is the transition from "It's quiet enough - leave it alone." to "Too loud - better turn it down." This is rather helpful because music does not have a steady volume. Without a knee, as the music crosses back and forth across the "trigger volume" (threshold), the contrast between unchanged and compress audio can be rather audible. Bigger numbers are softer transitions. Probably want it between 3-12. 6 should work well most of the time. * Now a series of "compander transfer function" values. These must be in pairs -except if one is missing it's presumed to be 0. Before I go further, I think I have to fully explain decibels. Uhm, well - I don't have time for that. So here's the short, what-it-means-here version. A "bel" is a measure of the change in a signal level. It's a _big_ change - so practically everyone uses "one-tenth-of-a-bell" called a decibel (dB). (remember your SI/metric prefixes?) So by definition, 0 dB means "same as", -6 dB means "6 decibels less", and 6 dB means "6 decibels more". In most digital systems, we measure everything relative to the maximum signal that can be described, which we call "0 dB FS" (FS standing for "Full Scale"). Now, any signal level can be described by a single value - say "-30 dB". This means that particular signal is 30 decibels less than the absolute maximum this digital system can have. The transfer function is really just a map. Each pair is telling the robot "If the input signal level is __, I want you to change it so it's __ instead." Effectively then a line is drawn between the point mapped out to give a final instruction set. So "-70,-60,-20" mean "When the signal is -70 dB turn it up to -60 dB, and when it's -20 dB turn it up to (the presumed) 0 dB." Often I set first pair to tie down the recording's silence. Since it's not really silent and I down want the noise turned up really loud, I start with a pair like "-90,-90". If it's from a cassette or VHS tape, it could be "-50,-50" The next pair is the the main relationship "-80,-20" is pretty extreme. "-80,-40" is medium. "-80,-60" is mild. For normal music (not special effects), you want each pair closer to zero than the previous pair. * Next is the "gain" value Because of the attack value, it takes time for the robot to turn down strong signals. This gain value is so there is sure to be room for the short-term signal spikes to not cause clipping. The faster the attack time, the less spare room you need here. If you get clips make this value a lower level. (which means a larger negative number). I find "-5" to often not be enough room and there's clipping, so I find I often have it set to "-10" or "-12". * A single number for the initial level value. This makes no change to the audio signal. Because of the attack value, it takes time for the robot to figure out what the signal level is and therefore what changes need to be made. If this value is given, it tells the robot "Pretend the signal level is __ when we start." This is especially helpful if the beginning of the sound file is _not_ silence. * Delay Put this many seconds delay between the meter and the volume control. In effect, this makes the meter able to see into the future. Usually a value the same as the attack works well. If you have a large decay, it can still work okay set longer than the attack. Too big, and the robot will be changing the volume at the wrong time. -- Philip |
From: robs <aq...@ya...> - 2009-08-29 09:23:21
|
--- On Sat, 29/8/09, Fmiser <fm...@gm...> wrote: > It is highly subjective - so "better" is tricky to > achieve... Wow - great reply, much better than my quick attempt :) |
From: Wesley C. <we...@se...> - 2009-08-29 13:54:56
|
On Aug 29, 2009, at 5:23 AM, robs wrote: > --- On Sat, 29/8/09, Fmiser <fm...@gm...> wrote: >> It is highly subjective - so "better" is tricky to >> achieve... > > Wow - great reply, much better than my quick attempt :) > Yes, thank you so much for the explanation, Philip! Going to digest it a bit and come back with questions. I find it interesting that when I play these recordings on my stereo system, it sounds great and the clapping doesn't bother me. But when I play it with headphones from my iPod, I find myself constantly adjusting the volume. Is this typical? I'd read somewhere that modern recordings are now extremely compressed because audio engineers assume people are listening with headphones. Wes |
From: Fmiser <fm...@gm...> - 2009-08-29 17:23:18
|
> robs wrote: > --- On Sat, 29/8/09, Fmiser <fm...@gm...> wrote: > > It is highly subjective - so "better" is tricky to > > achieve... > > Wow - great reply, much better than my quick attempt :) Thanks. I am an audio pro, and I do sometimes do training. Since I'm not much good at programming, I have to find other ways to contribute. Anyway - I hereby grant you permission to use any or all of it in the manual or soxexam. Edit as you see fit - no attribution required. -- Philip |
From: Fmiser <fm...@gm...> - 2009-08-29 17:32:46
|
> Wesley Chow wrote: > > On Aug 29, 2009, at 5:23 AM, robs wrote: > > > --- On Sat, 29/8/09, Fmiser <fm...@gm...> wrote: > >> It is highly subjective - so "better" is tricky to > >> achieve... > > > > Wow - great reply, much better than my quick attempt :) > > > > Yes, thank you so much for the explanation, Philip! Going to > digest it a bit and come back with questions. Your welcome. > I find it interesting that when I play these recordings on my > stereo system, it sounds great and the clapping doesn't bother > me. But when I play it with headphones from my iPod, I find > myself constantly adjusting the volume. Is this typical? I find it's not uncommon. Some of it has to do with the difference between "sound in your ears" and "sound in the room". Also, headphones and earbuds usually have _very_ good transient response (ability to reproduces quick signal spikes) compared to speakers. This is partly because of the mass of the driver its self, but also because of the damping effect of the air. When the driver is on your ear, it only has to energizing a 3 cm (1 in) column of air in a tube (ear canal) vs. the speaker having to energize at least 2 m (6 ft) of free air. > I'd read somewhere that modern recordings are now extremely > compressed because audio engineers assume people are listening > with headphones. Much of current pop music is very compressed. Partly for iPod listening - but also so it sound loud compared to the song played before and after it. Radio play is still a big factor in the economics of pop music. And humans almost always choose the perceptively louder song as sounding "better". Oh, another tip. You can use the sox command "play" rather than "sox" and hear the result immediately rather than processing a whole file and then playing it. play test.wav compand .2,.4 6:-90,-90,-80,-40,-10,0 -10 -30 .2 -- Philip |
From: Michael C. <ch...@mc...> - 2009-08-31 12:06:59
|
On Saturday 29 August 2009 5:30 pm, Fmiser wrote: > > I'd read somewhere that modern recordings are now extremely > > compressed because audio engineers assume people are listening > > with headphones. > > Much of current pop music is very compressed. Partly for iPod > listening - but also so it sound loud compared to the song > played before and after it. Radio play is still a big factor in > the economics of pop music. And humans almost always choose the > perceptively louder song as sounding "better". "The Insane Increase ..." graphic (four vertical red bars) at <http://www.digido.com/level-practices-part2-includes-the-k-system.html> might be of interest. MC |
From: Pascal G. <evi...@gm...> - 2009-08-31 13:57:42
|
On Mon, Aug 31, 2009 at 7:18 AM, Michael Chapman <ch...@mc...>wrote: > On Saturday 29 August 2009 5:30 pm, Fmiser wrote: > > > I'd read somewhere that modern recordings are now extremely > > > compressed because audio engineers assume people are listening > > > with headphones. > > > > Much of current pop music is very compressed. Partly for iPod > > listening - but also so it sound loud compared to the song > > played before and after it. Radio play is still a big factor in > > the economics of pop music. And humans almost always choose the > > perceptively louder song as sounding "better". > > "The Insane Increase ..." graphic (four vertical red bars) at > <http://www.digido.com/level-practices-part2-includes-the-k-system.html> > might be of interest. > Hmm... your link throws a 404 on me... -Pascal -- Homepage (http://organact.mine.nu) Debian GNU/Linux (http://www.debian.org) LACIME: École de technologie supérieure (http://lacime.etsmtl.ca) |
From: Wesley C. <we...@se...> - 2009-08-31 15:08:52
|
> > "The Insane Increase ..." graphic (four vertical red bars) at > <http://www.digido.com/level-practices-part2-includes-the-k-system.html > > > might be of interest. > > Hmm... your link throws a 404 on me... > There's something wrong with the url (the page is fine)... not sure what it is. Here's a re-paste of the url, plus a tinyurl link just in case: http://www.digido.com/level-practices-part-2-includes-the-k-system.html http://tinyurl.com/d22v29 Wes |
From: Pascal G. <evi...@gm...> - 2009-08-31 15:11:37
|
On Mon, Aug 31, 2009 at 11:08 AM, Wesley Chow <we...@se...> wrote: > > > > "The Insane Increase ..." graphic (four vertical red bars) at > > <http://www.digido.com/level-practices-part2-includes-the-k-system.html > > > > > might be of interest. > > > > Hmm... your link throws a 404 on me... > > > > There's something wrong with the url (the page is fine)... not sure > what it is. Here's a re-paste of the url, plus a tinyurl link just in > case: > > http://www.digido.com/level-practices-part-2-includes-the-k-system.html > http://tinyurl.com/d22v29 > Oh, there was a missing dash between "part" and "2". Thanks, -Pascal -- Homepage (http://organact.mine.nu) Debian GNU/Linux (http://www.debian.org) LACIME: École de technologie supérieure (http://lacime.etsmtl.ca) |
From: Wesley C. <we...@se...> - 2009-08-31 15:14:46
|
On Aug 31, 2009, at 11:11 AM, Pascal Giard wrote: > On Mon, Aug 31, 2009 at 11:08 AM, Wesley Chow <we...@se...> > wrote: > > > > "The Insane Increase ..." graphic (four vertical red bars) at > > <http://www.digido.com/level-practices-part2-includes-the-k-system.html > > > > > might be of interest. > > > > Hmm... your link throws a 404 on me... > > > > There's something wrong with the url (the page is fine)... not sure > what it is. Here's a re-paste of the url, plus a tinyurl link just in > case: > > http://www.digido.com/level-practices-part-2-includes-the-k- > system.html > http://tinyurl.com/d22v29 > > Oh, there was a missing dash between "part" and "2". > That's odd that the original link gave you a 404. Mine said "the page can't be found, but here's the closet link...", and it appeared to be coming from digido.com, not my ISP. That actually struck me as sort of a cool feature of the site. Wes |
From: Michael C. <ch...@mc...> - 2009-08-31 16:37:22
|
On Monday 31 August 2009 3:14 pm, Wesley Chow wrote: > On Aug 31, 2009, at 11:11 AM, Pascal Giard wrote: > > Oh, there was a missing dash between "part" and "2". Thanks for fishing me out of the soup. I only had it saved as a screenshot and keyed in by hand (and did check ;-)> For some reason SoX-list takes tens of minutes to return a message, for me, so I left checking it to later ... sorry. > That's odd that the original link gave you a 404. Mine said "the page > can't be found, but here's the closet link...", and it appeared to be > coming from digido.com, not my ISP. That actually struck me as sort of > a cool feature of the site. AFAIK it's a standard (optional) feature on Apache. Cool, but: -the usual if you type "amazon" you're probably looking to buy a book not for a naked mud-wrestling team. (NB Just noticed your "closet link" ... a lesser gaff than mine, but shows the problems.) -it perpetuates/propagates bad URLs. But glad it rescued my bad keying ;-)> MC |
From: robs <aq...@ya...> - 2009-08-29 09:16:33
|
--- On Sat, 29/8/09, Wesley Chow <we...@se...> wrote: > I've got some recordings of concerts to which I'd like to > apply range > compression. I've found that if I just increase the volume > on the > recording, drums and clapping tend to blow out the ear > drums while > listening with headphones... until I found sox's handy > compand command! > > compand 0.3,1 6:-70,-60,-20 -5 -90 0.2 > > Can someone suggest a better set of values to try? The numbers in the manual were chosen as a reasonable typical set, but have not been extensively tested by me at least so if someone has worked out a better set then please post them and I'll update the manual. Looks like the manual currently does an okay job of describing the parameters but here a few additional comments: The compander basically just turns a volume control up and down automatically. The first 2 numbers (0.3,1) control how fast the volume is turned down and up respectively -- the numbers are measured in seconds, so the smaller the number, the faster the volume is changed. Changing the volume too quickly may become obvious audibly so there is a compromise to be had here. The next section (6:-70,-60,-20) specifies what is known as the 'transfer function' and defines, for each possible volume in the input file, what you would like the output volume to be -- they're basically points on a graph and, if you have access to any of the gnuplot, Octave, or Matlab programs, you can see what this graph looks like using SoX's --plot option. The best way to evaluate audibly whether the transfer function is optimal is to check the output level from sections of the input audio where the volume is fairly constant. HTH, Rob |