I was trying hard to tidy my post up before you responded. Looks like you got it before my changes.

Personally, I don't really see any disctinction between the terms prevelant and loud in this situation. Perhaps you can clue me in. But before you do, please take into account that when it comes to human hearing, "loudest" in no way means the same thing as finding the sample with the greatest amplitude. You may want to review what is at the RG site in case you're drawing off your memory. I did.

I'm glad you did catch my point and I'm happy to be challenged when my words are incorrect and/or misleading.

EDIT:

I'm game if you change "perceived average level" to "perceived volume" and "loudest track" to "loudest perceived track":

I think including language that encompases silent tracks gets us back into the realm of minutia, though I added my own minutia with "loudest peceived track" though I think this edit removes even the slightest possibility that the statement still suggests that RG does something with peak samples.

Well, loud and prevalent don't have a related meaning to me. "Loudest" however is a superlative, which to me means that you'd be sorting the samples by their level. Prevalence implies sorting by number of occurences. I did indeed look at the rg site, since I got confused some 20 posts ago already
The page I think this whole thing is contained in is http://replaygain.hydrogenaudio.org/statistical_process.html
In my understanding, the sorting that is done there is by number of occurences and not by level.


Well I'm not a native speaker of English, so I'm easily mislead


All happy here

Well here's something. I'd like to see if anyone can provide some real-world examples (samples) of replaygained tracks that clip when decoded [in a player with the prevent clipping option off]. I have no idea what that would sound like...

Then I'd like to compare with what they sound like with clipping prevention on. I'm verrrrry curious. Can anyone help?

ryran: the fastest way to produce something like that is to slide the preamp setting up a real lot in your player.

I don't think so. I'm pretty sure the sorting is done by increasing level and the value is chosen 5% down from the top.

The easiest way to understand this is to look at the speech sample. At this point it should beomce clear that the levels are not ranked by occurrence.

As far as finding samples, if you use MP3Gain, just pick a track that the program indicates as clipping after the adjustment is made.

EDIT: Well my matlab skills are long gone so perhaps there is something here that rings a bell:

Pehaps you're right.

Aha, so you see... this is no case of nitpicking - we really have a different idea of the general principles in mind.

The speech example was what actually made me think the ranking was by occurrence. I thought everything above the 95% line (when ordered in occurrence) would be low-level background noise - that's the peak at -45dB or so in that graph. And then, taking the value 5% down gets you the less occurring value of -18dB. Makes sense, right?

But.... that was badly understood!
From http://replaygain.hydrogenaudio.org/mfiles/replaygain.m: "2) Sort in ascending order of loudness"

Thanks for all the patience... and sorry for confusing everyone if I did

(and I must add... somehow the ordering by occurrence still makes good sense to me...) ...EDIT: after some more thought, it doesn't make such good sense at all. If the distribution has a single maximum, and you start going down to the 95th percentile, you start jumping back and forth between the left and right flank of the peak.... not a very stable procedure. So forget my bad plan




I'm not

The thing that confused me there was that I thought the Vrms_all list was a binning list, containing "occupation numbers" for level bins... but it isn't.

@ Mike Giacomelli


i have wavegainfrontend but don't know how to adjust the desired level(i can't find this option),and i don't want to install Foobar only for this target.(and i have lots of audio editors working)
all i need is adjust album gain to 70%(for example), or is impossible?


thanks so much

I can put it quite succinctly then.
- I have no problem with attenuation (although I would prefer 'no change' over 0.000001dB attenuation. Less math is better.).
- I have a problem with limiting, and currently replaygain will limit some tracks and/or some peaks.

To make replaygain 'audiophile' like it claims to be, there needs to be a 'never clip, never limit' option in there. Which equates to peak normalisation, or attenuation only.

With the original 83dB target level, the only track I found which clipped was that of a single cannon shot (or firework? I forget!) recorded at close range with about a minute of low background around this single loud bang. I tried quite a few recordings - pop, classical, Jazz, "audiophile" demos, SQAM test disc etc. Nothing else clipped.

I think guruboolez found some classical recordings which could be made to clip. Frank Klemm also found some, but maybe by this time the reference level had been raised to 89dB.

You can certainly make foobar do what you want, as I've already explained (quoted above). If you really want to avoid scaling tracks which would only be scaled by a few dB, then you have to use some conditional commands scripted in foobar2k (which are well explained on HA, but I've never used them myself). You'd be much better advised to get a 24-bit signal path working and then stop worrying about it.

IMO opinion it would be preferable to do the clipping prevention on a per-album basis. There was an old Musepack plug-in which gave this option. I don't remember if foobar2k can or does do this.


Finally, just peak normalising everything on a per track basis is (IMO) possibly the stupidest thing that anyone could do (well, there's starting a war, using the old Xing mp3 encoder at 64kbps, having unprotected sex with a prostitute etc, but within the scope of the present discussion it's the stupidest thing you could do IMO), but you can do that in fb2k if you want to!

Cheers,
David.

It doesn't. There already is. But it doesn't equate to peak normalisation.

At the risk of sounding really grumpy (sorry!), how ReplayGain works (95th percentile, sorted by loudness/order/what day of the week it is) is irrelevant to the current discussion.

It tries to make all your audio files sound about the same loudness, matching them on a per-track or per album basis. Since there's rarely "room" to make quieter files louder, it has to make louder files quieter instead.

That's it.


You only need to worry about the frequency weighting, RMS window length, statistical sorting, and calibration if you're going to improve ReplayGain itself. I didn't claim any copy right on the idea, or how it's currently implemented, so anyone is very welcome to improve it.


I'd encourage anyone who wants to understand the practical implications of ReplayGain to play with it. The foobar2k implementation is really nice. I'm not trying to sell it to anyone. What I'm trying to politely suggest is that, like a lot of internet discussions, people could learnt a lot more by reading about something then going away and trying it than by sitting on a message board talking about it all day!

(Note: must apply lesson to self!) :-)

Cheers,
David.

Ow, that does sound grumpy

While I apologize for polluting the thread with a discussion that's indeed not relevant for using replaygain, I must say I got a better understanding from this discussion.

I'm a happy user and your tuning of the procedure works fine for my music too, so I don't think it needs improvement. But I still like to know how it works...

One more question: is it correct that the only tool available in linux to calculate replaygain values is vorbisgain? (I mean, I couldn't find a tool that handles more formats)
edit: ah, right... thanks iGold! ...I was actually (subconsciously? ) aware of those, so I don't know why I asked the silly question that way A nice thing about fb2k is that it handles several formats at once, that's what I was mostly thinking.

There is mp3gain for linux. Also wvgain from wavpack package. metaflac for gaining flac files... And so on

That's OK! The most interesting threads are those which go off topic!


I just thought some of the reasoning was a little bizarre. "I won't use it because my misunderstanding of how it works means that it can't possibly work". I didn't just jump in because it was ReplayGain - I have done and would do the same with various aspects of mp3 etc as well. How many "well meaning" people disable the low pass filter, disable joint stereo, even (gulp!) disable the psychoacoustic model to "improve" the quality of encoding!?

Cheers,
David.

.. which would be Bad™, I don't want it on per-song-basis

Ah, now we're getting some new info here, thanks. But this brings us into another discussion altogether, which is that perhaps now we're starting to attenuate too much because 'there are some tracks in the world that could clip at 83.5'. (I assume you mean "83.5dB above the noise floor, correct?) This is fairly poor, sadly.


Yeah, I 'd agree were it not for my personal situation which is that my professional DAC does not handle 24bit.

Well, for one, due to personal reasons (remote control compatibility!) Im stuck with winamp.

But secondly maybe we are losing track of what (I think) the goal should be, which is:

Make the player approximate what I would do myself, if I would sit behind the controls with all info (like track RMS, track peak, album RMS and album peak) available.

I know this seems as a bit of a drift from our discussion, but you see, in the end, if I were the audiophile DJ, I would:

1/ Try to make a track as loud as possible without any clipping. (i.e. lift it above the noise floor as much as possible)
2/ Probably use a goal RMS like 90dB above which I would attenuate instead.
3/ Not touch the controls (bit-identical passthrough) if no substantial gains can be made (without clipping)
4/ (of course) albumgain, not trackgain.

Actually, upon re-reading, all I would need is album RMS and album peak, and some sensible treshold for not touching the dials

You see, the difference between what you suggest (setting replaygain at 83dB) and what I would do manually is that I would aim for a certain perceived loudness, often attenuating, sometimes amplifying, and when necessary 'tolerate' the fact that certain albums simply can't reach the intended perceived loudness level since they have a few extreme peaks. (and go for 0dB peak level)


Of course, I would set it on a per-album basis! As I mentioned before, for albums that drop below the RMS, I would match RMS, and if that would cause clipping, go for (album) peak level.



I won't feel offended here, cause I believe I've understood the basic principles of replaygain since the start of this discussion. Maybe Ive been not making myself as clear as possible, but all I'm trying to do is suggest an alternative, truely audiophile setting for replaygain that makes as little compromise as possible while still improving the original problem (widely varying album RMS'es). Next up is teaching myself to code a replaygain plugin I guess, but perhaps I can persuade some current replaygain coder of the validity of my arguments

Care to explain why? Anyway, it doesn't happen on a per-song basis if you use album-gain mode.

No, 83dB SPL through a system calibrated to SMPTE RP 200. You could do worse than read the relevant bits in http://www.replaygain.org !

(In the digital world, it's a -20dB FS RMS pink noise signal, but this signal has to be played "in the real world" for real humans to hear it and perceive its loudness - which is the calibration point for ReplayGain. Simply matching the RMS of a signal to -20dB does not do the job of equalising loudness.)

Anyway, all implementations now use +6dB = 89dB (and implicitly, -12dB FS RMS equiv, but that is not a calibration point).

If you think -12dB FS RMS is "quiet" (in the digital domain), consider that any music which has a higher RMS (e.g. most modern pop music) has to be dynamically compressed to get the average up so high. Modern pop music is typically 6dB higher - to get there, it has to be heavily compressed and peak limited.


No, I think this is the interesting bit!



Why? You probably have 16-bit playback. If it's very good, the noise floor may be at -90dB. If not, then you could, in theory, get more music about the noise floor by increasing the gain.

However, if you process the signal, you have to re-dither it. That's 6dB of noise added, so (very simply and inaccurately) unless you're boosting it by more than that, you're probably not gaining anything.

Thus the only CDs you will process are those where the highest peak on the entire disc is lower than 50% amplitude!

If you ever find such a CD, let me know.



Which is exactly what you will get if you

• set the foobar2k pre-amp to +1dB (giving 90dB)
• write a script to switch off ReplayGain or erase the ReplayGain values for files with small ReplayGain adjustments
• switch on "clipping prevention by album" (if it's not a feature, it should be!)

Cheers,
David.

That doesn't make it any better.

Does that setting still work on a per-song basis if album-mode is set (would be funny)? The way I see it, if this uses the album-peak value, you've got exactly what you were looking for.

You are right!

In track mode foobar2k uses track peak. In album mode, it uses album peak.

I said it was a nice implementation!

Cheers,
David.

That's a good point, if I would do the digital mogrifications without upconverting to 24bits first. While I admit that in my specific case (I have to downconvert to 16bits again) you are right about this, in general, this would not be much of a problem.


A-ha. and there we arrive at the core of my request.. It is not a feature (yet), at least not in winamp in_mpg123, I do not know exactly about foobar2k.

[quote name='puntloos' date='Aug 23 2006, 17:18' post='423680']
That's a good point, if I would do the digital mogrifications without upconverting to 24bits first. While I admit that in my specific case (I have to downconvert to 16bits again) you are right about this, in general, this would not be much of a problem.
[/quote]

But if you had a 24-bit DAC with even a 19-bit equivalent noise floor, there would be no point to doing this.

The noise in the original recording is in the 16th bit (or well above!). With noise shaped dither, and a silent input signal, it might get down to the equivalent of the 18th bit. The noise floor of your (good, 24-bit) DAC is (hypothetically) below this already - what possible point is there in amplifying the signal in the digital domain?!

Just buy a good DAC (if you can).

[quote]
Which is exactly what you will get if you

• set the foobar2k pre-amp to +1dB (giving 90dB)
• write a script to switch off ReplayGain or erase the ReplayGain values for files with small ReplayGain adjustments
• switch on "clipping prevention by album" (if it's not a feature, it should be!)

[/quote]

A-ha. and there we arrive at the core of my request.. It is not a feature (yet), at least not in winamp in_mpg123, I do not know exactly about foobar2k.
[/quote]

It is in fb2k, as discussed above.

But hang on a second - you're using mp3s as a source, and worrying bit-perfect reproduction and noise at the least significant bit?!

Priorities, man! (or woman - never can tell, but usually safe to assume on HA)

Cheers,
David.
P.S. I wonder if your remote control can control foobar2k?

Btw I appreciate the time you're taking to reply

On with the show. Your current post is a bit broken but Ill modify


Yup, well from my DAC specs my noise floor should be around 103dB a-weighted, it's 20bits. See other topic, bottomline: I can feed my DAC 22k/24 and it should truncate to 20.


Sadly, the remote control app I use only handles winamp. There is a buggy & beta freeware alternative for foobar, and a mature and costly alternative for 'netremote'. so far I ain't switching

man. and well no Im definately not only using mp3s, its mainly monkey's audio or mp4. But replaygain and bitdepth settings all need to be set individually with winamp (since they are standalone plugins)

Oh I wish I wish I could get a proper remote for foobar that isnt $80 worth in software.

Have you looked at foobar in detail, posted a thread on the subject in that forum? I seem to remember a plugin for 8.3 that emulated a winamp interface so that a lot of remote software would work with it... Also, because foobar has an easy method to bind global hotkeys, if your remote device can work with any sort of configurable software (girder, uICE) it should be doable that way.

At this point, all we can say is "so sorry". There is software that has a full and perfect implementation of ReplayGain, but you can't switch to it. The software you want to use might not do everything just as you want... Which *we* can't really do anything about. ReplayGain makes songs quieter, that's just the way it works; if that makes the system as a whole too quiet than you have to find a workaround or just not use it. So sorry.
There does not exist any software that always does what you would do yourself -- the only software that ever does that is software that you write yourself. Which is why people write software, even when someone else has done something similar before. If you can't write software yourself, like many of us, you just have to live with the consequences of tradeoffs and find the software that best fits your needs.

...
I'd be willing to bet that you'll never hear clipping even if mpg123 doesn't read the peak tag to prevent it.

Or use alternative input plugins for WinAmp.

Say, IIRC the latest (still beta, though) release of WinAmp can do a ReplayGain scan.

FWIW I have some 900 CDs worth of FLAC files on a hard drive (old and new rock, jazz, classical), all of them a played with foobar 2K using album replaygain, with clipping detection on but NOT using clip correction, and thus I can report that the number of time RPG sends playback into clipping (i.e, pops up the 'clipping detected' window) is relatively *miniscule*. It's not a problem worth being anal about IME.

Wrong again. No, it is still not a limiter. Have you tried switching on "clipping prevention?" It scales down the tracks if needed.

It's probably been proposed somewhere before, but I wonder if replaygain (e.g. in foobar) should offer the option of only ever *trimming* level where indicated (negative RPG values) , but never boosting level (ignoring postiive RPG values). The vast majority of rock/pop releases seem to require trimming rather than boosting anyway. This would also avoid the issue that occurs when using track gain (rather than album gain) , where something like the quiet opening track of "Dark Side of the Moon" gets boosted to ridiculous levels.

isn't it the boost/trimming relative to the whole track (or album if in album mode)? i don't understand what you say.

FLAC files do not clip!

LOL

I enjoy this ressurection for this one post.

Nor I, you.




Replaygain can boost the level of a track, or more rarely (at least for rock) a whole album. In either case that *can* result in clipping. In the case of track gain it can also mean that a track that's mostly quiet (like the opening track of Dark Side) will sound bizarrely loud. (I suppose that could also happen with album gain, if the whole album is very quiet).

The boosting or trim is relative to its 'native' (recorded) level. Zero boost or trim means the track (or album) is output at the level it was recorded. The 'target' level achived by boost or trim is built-in to replaygain -- it's 89 dB SPL (the RMS value after virtual application of 'equal loudness filters' -- these filters are not applied to the actual output, they're just used to calculate 'equality' in a psychoacoustics-approved fashion)


("FLAC files do not clip?" Someone's been hitting the crack pipe.)

What was probably meant was: "Encoding with FLAC and other lossless encoders will never change, add or remove clipping; unlike what can sometimes occur with some lossy encoders.".

-brendan

It doesn't make it any less funny!

i hope you know that this was intended

oops!

-brendan

Replay gain is very nice. With replaygain tags I use 0dB preamp, without tags -10dB. That ensures that (most) tracks without replaygain info don't sound louder than tracks with RG info.