Skip to main content ↑ ↓

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: Regarding MP3Gain & clipping, decibel settings... (Read 9631 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Regarding MP3Gain & clipping, decibel settings...

Ok, for some time now I've been using MP3Gain (or, specifically, MP3GainGUI on top of MP3Gain) to normalize my MP3s (yes, I normalize.  I find that using the RMS "maximizing" method seems somehow to not work as well for my ears for whatever reason, so I stick to plain peak normalizing.)  One thing I've been unclear on almost from day one is what to normalize to.  The program defaults to 89dB (eg trying to ensure it stays just under 90dB) and most things go by this.  ReplayGain seems to make this goal basically unoptional in fact.  I'm not really clear on exactly why we pick this number though.  One old post in another forum I found via googling I read said that 6dB will be added by the decoder or something along those lines, but that modern decoders don't do this?  Of course, I take any random post found via google with a grain of salt (actually, preferably two as they can be very hard to swallow sometimes.)  It would explain why that target of just under 90dB though as I've read that 16-bit has a limit of 96dB.

There's just one thing.  I keep finding CDs that actually scan as over 100dB!  If 16-bit is limited to 96dB, how'd they even get there?  Bad ripping?  I must admit that even I have a few bad rips from my past before I understood more about how the CDDA format works and before I discovered EAC (not to mention back when I had cheap audio equipment and believed that crap about 128Kbps being CD quality, lol.)  Sadly, some of my old stuff can't be saved beyond just trying to clean it up as much as possible and leaving it alone as I have lost many things during the process of several moves in the past few years (with one more coming up...  This time I'm making quality backups, lol.)  What I find especially interesting is that, while some of these show up as detected clipping for obvious reasons, I've seen one or two that did NOT say they detected clipping...  Needless to say, this has only succeeded in further confusing me in this matter.  Is it just not detecting well enough?  But, to really throw things off, I bought a game on GoG and downloaded its soundtrack only to find that, even though the soundtrack had a peak below 90dB, clipping was detected in several tracks...  Should I just assume the clipping detection isn't accurate enough and completely ignore it now?  Unfortunately, I truly hate clipping.  It sounds horrible when I listen with my headphones especially.  At times downright painful even.  So I still want to avoid real clipping regardless of detection.

One reason I've been dissatisfied to just leave it at < 90dB and forget about it is the 90dB limit is being something of a problem for me at times.  For one, I'm having to almost completely max out my amp with this.  Whenever I listen to something I haven't processed yet, it nearly blows my eardrums due to the as much as 10dB difference at times.  Worse, I listen on a portable DAP whenever on the go and I'm finding that I'm pretty much having to max it out and sometimes this still isn't quite enough.  (It doesn't help that if I want to use my headphones without an external amp, they need a pretty good bit of juice and only my D2 can handle them and even then at a fairly high volume setting.)  To this end, I did a little more digging around and found where one person said that 92dB might be an ok setting for mixed music (mine is definitely pretty mixed) and I've been trying this with so far no obvious signs of clipping.  Perhaps there is clipping and it's in too few frames for me to really hear, but I guess that's fine.  (Well, except that one soundtrack which had clipping even at lower numbers, but I didn't end up keeping anything from that soundtrack anyway.)  92dB definitely seems to strike a better balance between being a decent volume and avoiding clipping so far for me at least.  Unfortunately, there seems to be no way to get ReplayGain to aim for something like this as it seems to be aiming lower apparently.  (When I left MP3Gain on 89dB, it would add only just enough to reach 90dB, but when I started going to 92dB it started subtracting instead.)  I probably use MP3s more than FLAC or Vorbis anyway and sadly my DAPs completely ignore ReplayGain.  )-:  Well, on my PC I'm using 24-bit precision, but I notice hear clipping on my DAPs so far at least.

The main thing I'm asking here I guess is just why exactly we aim for a number like 90dB and why so many things can be over 100dB with some not even reporting clipping.

  

Regarding MP3Gain & clipping, decibel settings...

Reply #1 –
ReplayGain works by estimating the loudness of the music track as a whole, then comparing the loudness to that of a "reference" sound (pink noise reproduced at 83dB SPL). The final value is the amount of attenuation/gain necessary to reproduce the track at a nominal loudness equivalent to a sample of pink noise reproduced at -20db RMS, assuming it is played back at 83db SPL. 83db was chosen because it is a film audio standard.

All extant RG scanners tack 6db onto the correction value, making the reference loudness effectively equivalent to 89db SPL, because 83db tends to attenuate most music far more than many people are comfortable with. Most RG decoders like foobar2000 can undo this by a gain factor applied to RG-processed tracks.

The relatively low loudness of pink noise, and its relatively high peak-to-average ratio, means that music mastered today can easily be 15db greater in magnitude from the RMS measurement alone, and possibly up to 10db louder for psychoacoustic reasons. So that's why the number can be above 100db.

Regarding MP3Gain & clipping, decibel settings...

Reply #2 –
The 89dB number is supposed to be SPL (acoustic Sound Pressure Level).

In reality, the correlation between the decibel level in the digital file and SPL depends on a bunch of un-calibrated stuff...  The volume control setting, the gain of the amplifier, the efficiency of the speakers, the listener's distance from the speakers, etc.    You can't really say, "This CD goes to 100dB", because any CD can go to 100dB with the volume turned-up.  (Movie theaters are calibrated, so you can tell from the digital file exactly how loud it will be in the theater.)

SPL is referenced to the threshold of hearing. 0dB is the quietest sound and the dB SPL numbers are always positive.  The dB level in a digital file is usually referenced to the "digital maximum" designated as 0dBFS (zero decibels full scale).    So, 0dB in a digital file is the maximum and the numbers are almost always negative.  (There are exceptions, depending on the software/format.)

The creators of ReplayGain & MP3 Gain have simply chosen a calibration correlation between the acoustic level and the digital level.  I can't find my notes at the moment, but I think it's something like 89dB SPL = -6dBFS.

They had to choose some correlation level because of the Fletcher-Munson equal loudness curves, and they had to choose something higher-level from theater calibration, because theater calibration leaves too much "headroom" and this would make your MP3s too quiet on the average player/system.







Regarding MP3Gain & clipping, decibel settings...

Reply #3 –
Unfortunately, it still tends to come out too quiet.  )-:  I'll admit my biggest problem of all has been the car sound system.  Most sound systems seem to not quite amplify the line-in as much as they should perhaps.  With 92dB, I'm mostly pretty satisfied with the results on my D2, though I'm not going to say that there's absolutely no room for improvement either.

Anyway, while I hope to understand the process better, the key thing I'm trying to get here is an idea of what sort of number to use in my normalizing.  After reading that one guide I've been going with 92dB, but generally speaking, the higher it can go without me detecting any clipping (and I guess all that matters here is whether or not I detect it rather than whether or not an algorithm does,) the better.  I need something I can apply somewhat universally to my audio though.  It takes a very long time to process EVERYTHING though (and even longer to compare) so I'd rather go about this more intelligently rather than simply trying to brute force find that ideal number.  Is 92 about as ideal as it's going to get for a higher number than the 89 original target?

BTW, is there anything besides ReplayGain that can be done for FLACs and OGG Vorbis?  It's kind of annoying how my DAPs won't support it.  Hopefully someday DAP designers will think of such a basic feature given how long ReplayGain has been an accepted standard now, but, in the meantime I have only the more "brute force" methods like MP3Gain.  Well, FLACs I can modify myself since they are lossless (though it would be nice to have some way to be sure I'm getting them at more or less the same peak values so I don't have to do quite as much volume adjustment between songs,) but, I can do nothing for OGG Vorbis.
  
 

Regarding MP3Gain & clipping, decibel settings...

Reply #6 –
Thanks.  I never would have thought of that...

Ok, seriously, give me a little credit here.  Of COURSE I turned the volume up...  My headphone amp nearly has to go to the max and my D2 is all the way at the max.  The problem is, these headphones need a decent amount of power to do their best due to that low impedance for full sized, full range drivers.  When the music is at its original volume level, these things have no trouble and, in fact, I'm having to do less actual amplification by far, but with everything run through MP3Gain it gets more tricky.  The D2 is struggling as it is with these headphones with the music set to 92dB and a few songs were just not right at all when I had them on 89dB with most being just a tiny bit too quiet so I HAD to use an external amp just to listen.  This is without an impedance adapter I might add.

Oh, and I'm in the US, not Europe, so there are no arbitrary volume limiters being applied to any of my stuff.  I don't even want to THINK about what this would be like on an iPod with its probably much weaker output.  (From some quick googling, it looks like most iPods have 30mW per channel, so it's perhaps not a lot less, but, given that the at least 37mW per channel that the D2 provides is a bit too quiet, I suspect that reducing it a further 7mW would NOT be a step up...  It also sounds like they don't have the strongest amp out there, perhaps since Apple probably intends them to drive those junky earbuds that come with the iPods more than anything else.  The D2's DAC/amp definitely has a little heft to it other than the slight bass rolloff annoyance at low impedances.)

Remember.  I'm talking about headphones, not IEMs or earbuds.
Quote
This was covered a little while ago, hopefully this might help:
http://www.hydrogenaudio.org/forums/index....st&p=633846

C.

It does.  Looks like what's being said there is pretty similar to what I found with Google's help too.  Honestly, I don't want to have different songs set to different levels to avoid clipping, so I guess I'll have to stick to 92dB as a reasonably safe maximum as even that is pushing it a bit.  Well, it's pretty close to perfect most of the time anyway even with my headphones on the D2, but it's a bit annoying that I can't use an impedance adapter with my headphones to solve the bass rolloff without the music as close to as maxed as possible.  I can live with just a smidge less bass in the absolute lowest ranges though anyway.  (Maybe I'll use the SPL-based gain -- what I was essentially referring to as RMS earlier for simplicity's sake though maybe not as accurate? -- for a few songs.  I'll have to figure that out later I guess.)  That extra 3dB of difference at least helps a lot.  In the car it's just fine in fact.

Thanks to everyone.  With all of this I do understand a bit more of how it works and, more importantly, what sort of number to shoot for when normalizing.
  

Regarding MP3Gain & clipping, decibel settings...

Reply #7 –
I'm a little suspicious that you are having volume issues with a 106 dB/mW headphone, with an external amp. I've run my Audio Technica A900s unamped from my iPod, and it's more than loud enough for any safe listening situation. And that's a much lower output combo than what you are running. I mean, seriously... your headphones do NOT need "a decent amount of power". Your headphones are higher sensitivity than my ATs and are in fact, equal in sensitivity to ER-4Ps.

92db is fine as long as you enable "prevent clipping" and do ReplayGain scanning after lossy encoding. It's going to mean that anything which nominally RGs to -3db or above, and peaks at 0dbFS, is not going to sound as loud as it otherwise would. That is very rarely an issue.

Besides that, buy a bigger headphone amp, buy more sensitive headphones, or get your hearing checked.

Regarding MP3Gain & clipping, decibel settings...

Reply #8 –
Quote from: Axon on
I'm a little suspicious that you are having volume issues with a 106 dB/mW headphone, with an external amp.


The main issue is actually WITHOUT the amp.  With the amp, I find it annoying that the volume level is so radically different between the files I've processed and those that I have not, and I wonder if it's really ideal to have it amplifying more if the sound could be coming out louder on the line level to begin with, but I can live with using the processing with the amp.  The amp is the iBasso D3 "Python" which can drive just about anything short of those insane "speakers on a headband" that AKG once came up with (seriously, they were basically speakers actually requiring whole watts, not milliwatts to drive them...)

As for the 106dB/mW, I have a suspicion that this isn't quite 100% accurate.  Upon reading some reviews I found them described as being less sensitive than many, and honestly I kind of have to agree.  These D2000s are actually requiring more volume and amplification than my previous D1001s despite the fact that Denon describes the 1001 as 103dB/mW versus the 2000 at 106dB/mW -- and I don't think it sounds like a 3dB difference either.  I suspect there's something else missing from the equation here really.  If anything, I shouldn't be able to notice a volume difference between the two...  Now, if we JUST go by sensitivity to judge here, my Ultimate Ears Super.Fi 5 Pro IEMs have an impedance of 21.8 ohms or so roughly (versus my recabled D2000s at 22.7) have a sensitivity of 116dB/mW and I'd have to say that they are considerably louder by comparison (not such a shocker given that they are IEMs after all) than a difference of about 9% would imply.  There must be some other factor here as the headphones definitely need more current to fully drive them than the IEMs, and not 9% more either.  Well, perhaps it's not as simple as this, I don't know, but either way, it's just not seeming to work out as simply as that to me at least.

Quote
92db is fine as long as you enable "prevent clipping" and do ReplayGain scanning after lossy encoding. It's going to mean that anything which nominally RGs to -3db or above, and peaks at 0dbFS, is not going to sound as loud as it otherwise would. That is very rarely an issue.


Well, I still find the ReplayGain to be annoying is the thing.  I have to have the volume almost all the way up with it, versus only most of the way up otherwise.  If nothing else, it's less painful when other sounds/songs play that haven't been similarly processed even if it is only a 3dB or so difference.


Oh, and yeah, I do enable the prevent clipping setting.  So far very few files have ended up not being able to go past 89dB this way.

Quote
Besides that, buy a bigger headphone amp


The headphone amp is fine, just unnecessary for the D2.  I need it for some other players, yes, but why should I have to have an extra box hanging off of my DAP that isn't needed and drain power from those batteries all the faster unnecessarily?  (And I only have enough batteries for one set to go in at a time, so when it runs down enough to start clipping I have to wait for them to charge again before I can use it again.  I'll buy more later, but, atm I have some none too fun expenses and I can only see this as an expenditure that should come AFTER food/etc.)  The amp works great on my PC and other players, but the D2 is capable of powering these headphones sufficiently not to need it for anything other than the bass rolloff (but I'd rather use the player as the bass rolloff isn't actually bugging me very much.)

Quote
buy more sensitive headphones


You're kidding, right?  You don't just change headphones because their sensitivity isn't as high as it could be or something...  Besides, I couldn't afford to if I wanted to right now.  Even if I could, I have these headphones not because they looked pretty or something because they have the right sort of sound for me.  What would you suggest I do then?  Get the D7000 for a 2dB/mW increase?  Anything else is going to be a significant change from an otherwise pretty ideal formula.  And for what?  Better DAP support?

Quote
get your hearing checked.

My hearing may not be perfect by any means -- in fact, I'm sure it's not.  However, it's certainly good enough and better than a lot of people I know.  It's DEFINITELY good enough when unprocessed files play and I have to scramble for the volume control...  Remember, some of these things are being lowered a good 10+dB even.

PS.  Again, just to be 100% clear, the sound is QUITE loud enough with the original sound as it originally comes before MP3Gain/ReplayGain processing -- in fact, often enough too loud, which is part of my problem since I have the volume way up for the processed files.  It is only after decreasing the volume that these songs start to not sound quite right at very high volume levels.  Of course, in case it's not clear, I'm trying to somewhat level out my whole collection, not just individual albums or something, so varying levels wouldn't really work even if some could be set higher than others with no clipping.  I hate having to fumble for the volume control while something too loud blasts away at my ears especially, but too quiet is also bad in many ways.  For instance, when I resort to listening to music on my A320 (which I have to do a lot at work since I can't fit it AND the MP3 player, but also want to read books and frankly the D2's text reading feature is lacking at best) the stupid thing insists on playing every sound file it can find and since I have Dingux and several games installed, it's also picking up the sound files from some of those games and playing them at times in my playlist as well (unfortunately, there seems to be no better option atm for handling music.  GMU in Dingux has issues -- namely skipping and etc even on MP3s, not to mention more complicated formats -- and more importantly, I listen to music while reading when I'm using the A320 but the text reader in Dingux is worthless and even if it weren't, the software isn't set up correctly for multitasking atm.)  Most of the time it's a minor annoyance, but, on occasion it has been downright painful with the sound volume set for the processed music's level.  Thus I had hoped to strike something of a better balance between the two.  Sadly, it sounds like if I go up any further I'm going to start getting a lot of clipping (or a lot of unprocessed files, which still defeats the purpose.)  Oh, and to be clear, it doesn't always even have to do with the headphones.  I find it annoying on the D2 that it can't quite drive them when I use processed files, but has no problems when they are unprocessed, but with the A320 I use my IEMs which it can normally drive well enough.  On the PC I use the amp for obvious reasons and it can definitely handle the headphones just fine.
  

Regarding MP3Gain & clipping, decibel settings...

Reply #9 –
I've tried driving my HD280 Pro headphones from an iPod touch, a Zune, a Sansa and the Cowon iAudio 7. I had to crank up the volume on the Touch, the Zune and the Sansa in order to be able to hear music. The Sansa player was probably the hardest to hear anything, since I was getting a lot of noise from the unit itself.  The iAudio 7 will drive my HD280's, but even that I have to crank it up more than with either the stock earbuds or my UE Superfi 3's.

All that being said,  maybe the easiest way to cut this Gordian knot would be to use your UE Superfi 5's on the D2 and not listen to some of the music you've had problems with in the car?

Regarding MP3Gain & clipping, decibel settings...

Reply #10 –
Using 92dB MP3Gain is good enough.  Not perfect, but, good enough.
  

Regarding MP3Gain & clipping, decibel settings...

Reply #12 –
Yeah, I know about the loudness war, and, unfortunately, the greater majority of my stuff has severe dynamics compression.