I was just wondering why Foobar2000 seems to be doing the replaygain so badly on my files. I'm trying to get the albums in my collection maximized and individual tracks on my DAP as well so I'm not having to always fiddle with the volume controls. For the most part, MP3Gain has solved this problem (especially for my DAP which, as usual, does not support replaygain) but not all files are compatible with it... (Most notably my FLAC files, but I have some OGG Vorbis files as well -- I've collected my music from numerous sources with some even just being files straight out of a game's music directory.) It doesn't do much good on my DAP, but I wouldn't mind if on my PC at least I didn't have to fiddle with the volume all of the time. I've actually had better luck completely disabling replaygain as most of my music is much closer to 0dB peaks than the replaygain would put it... For example, I have one song with a peak at about -0.5dB that Foobar2000 wants to add a -7.55dB gain to for some reason... I've seen one with a peak below 0dB that it actually wanted to apply something in the range of -11dB to even! I had the volume control on my speakers set all the way up to about 3/4 of the way up (and when the next song started that didn't have a -11dB bias, it was rather painful to say the least...) Setting it that far off is really bad (especially in light of the fact that decibels use a logarithmic scale and thus -11 is actually a HUGE difference...)

I'm guessing that it's one of those that try to use algorithms to determine what they think sound the same volume levels to the human ear (eg "maximum loudness" rather than "maximum peaks.") If so, it's not working very well... To my ears, after applying -7.55dB it sounds quieter than a quieter song does to me, though I can see how you would have to apply something like -7.55dB to it to get it to sound quieter and closer to the same volume level as a quiet song. Maybe my ears are just weird, but I really wonder if that's even possible. Regardless of what exactly is going on here, I don't like it and I think it even kind of defeats the purpose (after all, if my ears were really finding these songs to be at the same volume levels, would I be adjusting the volume by such extreme amounts? Maybe a little from time to time, but I'm finding that I always have to turn up the songs with a really negative gain and turn down those with a positive gain...) Is there anything I can use to get my FLAC and Vorbis files normalized by peaks rather than by any algorithms that think they know what I hear better than I do? Is Foobar2000 capable of doing this, or do I need to use something else (and if so, IS there something else?) I know I need to figure out something else for my DAP (not a problem with the lossless perhaps since I could just normalize them, but I can't do that with my OGG files) but it would be nice if I could properly use Foobar2000 for this. Especially since I'm using it to help analyze the music so I know what needs to be done for my DAP. (I'm a huge proponent of minimalism, but at this point my Foobar2000 is decked out with a bunch of visualizations to show me different aspects of the sound ranging from the simple VU meter with peaks set to not fall until a new song starts to a spectrum analyzer that shows me frequency usage.) If nothing else, by getting Foobar2000 to properly handle replaygain I can at least tell how far off something is when it comes to analyzation. But, more importantly, when I'm actually sitting down and listening at my PC I want it to be consistent...

PS. Don't misunderstand me. I'm not a fan of compression at all or anything and am sick of the whole loudness war. I just want music to be more normalized so that the only differences are in the dynamics. I suppose if I had more difference in types of music that I listened to (for example, if I had classical and rock albums both on there at the same time it could maybe be a problem) it would be an issue, but since I don't the huge differences between them turns into some being more limited than others as I hit maximum limitations of things like my DAP which just can't amplify some by enough for the IEMs I'm using (normal properly normalized music is just fine with a decent amount of leeway, it's just the songs that end up with really low peaks that are a problem.) Besides, the originals aren't actually set correctly. Sadly most of my music is modern enough to be well within the loudness wars and not really by independent artists for the most part, so the originals all would be closer to 0dB (or, more accurately, 89dB) anyway and it is more a mistake of the ripper that the files produced are not. (Well, hopefully someday I can get enough money to buy all originals -- maybe even vinyls -- so I can rip properly myself, but at this time I have to make do with using various sources that actually offer digital downloads rather than only physical CDs.)

In order to have Replaygain to work well, you will need to replaygained ALL of your files.

If you already do that and still not satisfied, well.. it's not perfect.

And IMO, Album gain work better than Track gain.

You seem to have some serious misunderstandings there:

- Replaygain is not peak-normalization
- peak-normalization is not loudness-normalization


As for RG being inaccurate.... try using albumgain always - even if the album is incomplete (i.e. if you only have a few tracks of it). I dont use trackgain at all simply because of it often beats the purpose of RG. However, even with albumgain, you WILL need to do some adjustments sometimes - especially since i noticed that on some albums, the tracks are already "off" on the CD.

foobar2000 implements the Replay Gain algorithm as described on the Replay Gain homepage. As you have guessed, Replay Gain is based on a loudness estimation rather than the (perceptually less relevant) peak values.

If you want to try if you like it better, you can nevertheless use Replay Gain in foobar2000 to achieve peak normalization:

• Use the Replay Gain scanner on your files to determine the peak (and gain) values.
• On the Playback preferences page, set the source mode to album or track, the processing option to "prevent clipping" and the preamp value for files with Replay Gain information to maximum.

Preemptive note: Lossy files may have peak values above 1.0 and may thus be played with reduced volume.

Another note: When using Replay Gain in the normal way (i.e. "apply gain") it is often helpful to set a negative preamp for files without Replay Gain information.

Yes, I did that. I'm aware of the fact that if they don't all have it it won't work right.


What did I misunderstand? I got those two points, so what is the misunderstanding? If replaygain can ONLY be used for loudness normalization, then I think it's going to be worthless for a lot of people and that we all need an equivalent method because you can't always reencode everything just to manually normalize it... A software based solution is perfect because there is no conceivable reason why the software can't just simply increase or decrease the volume as needed whereas having to reencode means a lot of quality thrown away due to software just not being up to the task (and in that case there is no reason it should stay outdated for so long.) Basically part of what I'm asking is if there is any way to make it set replaygain information based on peaks rather than loudness.


I did that. I just used trackgain in the example above to make the point that it was not adding a huge negative bias due to another track in the album.


That would explain it.


Won't this kind of half compress all of the sound though? It's already compressed enough as it is... (And I mean dynamics compression.)

ReplayGain doesn't compress anything.

Cheers, Slipstreem.

It can in the situation described. Just to be clear, as I said, I'm talking about dynamics compression. Eg this:
Original track peak normalized:
http://img443.imageshack.us/img443/2954/mi...rmalizedzm1.png
As you can see it's pretty bad. There isn't much dynamic there with huge sections basically just sitting on the maximum volume level (this is due to the loudness war I mentioned earlier and there is nothing I can do about it except try not to make it worse.)

This is what happens in a worst case scenario of the method described above if you maximize the amplification on the audio data and just use clipping prevention to prevent it from going over the limit:
http://img443.imageshack.us/img443/5846/mi...ximumampxz0.png

Not a pretty picture there... Where the original looked really bad with no real dynamics to it once the actual chorus starts, this has almost no dynamics left anywhere... The song is SUPPOSED to start out quiet with a bit of orchestral sound to it and this turns it into some sort of rock orchestra... The original is bad enough, but this is beyond unacceptable for me. Basically, it "compresses" the volume levels. Smaller volumes are raised by the amplification factor and larger volumes are shrunk by the clipping prevention so that the average loudness gets higher. Since this isn't even applied properly at the level of the master, it is much more inconsistent too, so you get an unpleasant sort of volume "shivering" I guess you might call it. (A sound engineer working on the master would be able to deal with instruments and s uch seperately rather than the entire thing together, so while they do still insist on compressing the dynamics, they can keep them consistent at least.)

I think you misunderstood this and thought about "clipping prevention" only as a post-limiter. But the method described above was to maximize the amplification, then use the peak as scanned from the original data to prevent clipping, so that the peak is +/- 1.0 on linear scale and no dynamic range is lost.

Thanks. I didn't misunderstand but, in fact, this is exactly what I was trying to ask about. (The post you quoted was in response to Slipstreem's post, not foosion's.) I think this answers my question.

You seem to do many conclusions based on wrong facts.

Replaygain will equal the percieved volume level. It does this by defining an acceptable volume level, by the standard defined to 89dB. This level is almost always beyond any songs average level, and therefore most songs will need to be lowered in volume (that's why you have to turn your amp a little more up than usual) - That's what you see when a replaygain value is as "extreme" as -8 and -10. If you take a look at early CD releases usually from '80s, you will see that most of these comply pretty well with the 89dB default value, and will only have small replaygain values.

I fail to see why you bring (dynamics) compression into this. Replaygain has nothing really to do with it. It will NOT compress your songs (you will need a dynamics compressor and limiter for that!!!). You need to realise what i described above to see what I mean.

I have an idea of why you have experienced some tracks that wound louder than others: If you with mp3gain have used a replaygain value greater than the default 89dB, your songs will not have the equal percieved loudness, and even so most of your mp3's are likely to be very distorted during playback.

If you DID set mp3's replaygain value greater than 89dB,. i suggest that you undo all these, and/or modifies all the mp3's not to clip (an option in mp3gain), or you might experience badly clipping in some songs.

When the percieved volume level is uneven, this always mean that the calculated levels are wrong - Youu will always be able to recalculate the replaygain values in foobar2000.

Edit/addendum:
You obviously have read up on the subject of dynamics compression - That's good, but don't confuse it with replaygain! Although the "gain"-part of the name could be somewhat confusing, it will almost never increase the volume.

In this case it very much seem like you have used MP3Gain and entered some insane value (100dB or something?) and applied it. The result is NOT dynamic compression, but extensive clipping.You can use replaygain for that purpose, but:It sounds like your DAP would not benefit much from it as it's not very good amplifying music with lower levels. In that case replaygain might be a little tricky, as you probably need to set a higher default level in MP3gain. Always be careful when you do this, and almost never go above the calculated DB level. Personally i find 94-96dB to be sufficient for newer music (it will just lower it a little). However this is most likely going to be a problem with foobar2000 calculated values, as it uses 89dB as default.No, from your descriptions you need replaygain - Not peak normalization!You should forget ALL about peaks now. You have a misunderstanding that peaks (or compressed music) would have the same percieved loudness level - THIS IS NOT TRUE! Dynamics compression can be applied from very subtle to almost white noise sound. Both can look very blocky in a wave editor, but they are percieved very differently.Congrats! This is exactly how replaygain works. Do this on all your songs, and you will be fine!Ummm.... NO?!? The ripper (or the way CD's are ripped) has nothing to do with volume, percieved volume, normalization or even dynamics compression. So FORGET that! Ripping CD's is a pure CD-to-harddisc "copy" and would (SHOULD!) never touch the audio!

As I said, ReplayGain doesn't compress anything. It just generates a gain offset value for ReplayGain-compatible players to apply in order to keep the perceived loudness of tracks/albums relatively constant from track-to-track/album-to-album. Any compression that you hear is a product of how the player performs once ReplayGain information has been applied.

I realise that this doesn't address your problem, but I wanted to make sure that you weren't under the false impression that ReplayGain was a dynamic compressor, because it isn't.

Cheers, Slipstreem.

Everything has been said about the topic, but maybe this page could clarify things for future readers:
http://eolindel.free.fr/foobar2000/Replaygain.php

Should this link http://eolindel.free.fr/foobar2000/Replaygain.php provided by Bachi-Bouzouk be used for the best results for playback of replay gain? According to foosion, you should use :


But what about if you're using advanced limiter dsp, do you then really need to set playback in source mode and processing mode, and if so go with foosions advice instead of the top link's way of doing this.

ActiveThoughts: Please read the thread again. I merely described a way to do peak normalization with the Replay gain functions in foobar2000. This is a rather unusual way to use Replay gain and definitely not the recommended or standard way to use it.

I am sorry Foosion, i got lost in translation somewheres. Could you tell me a good/or proper way of handling play back of such files with replay gain. I use Advanced Limiter alone right now.

First to quickly address the issue of formats where you can't apply ReplayGain to the files as you can with MP3 for using on your DAP...
If you want to play FLACs on your DAP with RG, you could convert them to MP3 with ReplayGain processing on (a safe transcode unless the FLACs were from lossy sources, such as DRM-protected lossy files) or simply use Foobar's Convert to convert to a lossless format with ReplayGain processing turned on - possibly setting the filename extension to .RG.flac to prevent you from confusing them with the original unadjusted FLACs or quarantining them into a specific folder for adjusted DAP output files. Or you could use Rockbox if available on your DAP, which supports ReplayGain.

It sounds possibly like you want to have loudness similar to modern over-loud undynamic CDs without applying any dynamic compression or clipping to the more dynamic (not so loud) CDs in your collection. Unless you have a replaygain implementation that adjusts the volume control you simply can't do it.

The alternative interpretation of what you want is to achieve peak-normalisation at the same time as matching perceived loudness (i.e. preventing the need to reach for the volume control). I don't know why you'd want to do this because peak values mean very little to humans, and with no distortion it's not possible.

Replaygain almost always reduces volume precisely because modern music releases are mastered too loud to allow good headroom for loud peaks, so bringing 1980s or 1990s releases up to modern average loudness would necessitate squashing those loud peaks or turning up the volume control, the latter option being unavailable in current practical implementations of RG.

Advanced Limiter is a look-ahead limiter so it won't cause any dynamic compression unless that signal is about to clip, in which case it applies soft limiting. This could work to some degree in that it won't further squash the dynamics of modern CDs yet it will squash the dynamics of older/quieter CDs. You ought to turn off clipping prevention if using Adv Limiter.

You could set a RG preamp of 3-5 dB to aim for about 92-94 dB loudness and turn-off clipping prevention. This would probably avoid the grossest distortions and come a little closer to the loudness you're after without necessitating a higher volume-control setting.

If your DAP isn't loud enough to overcome noise (such as when commuting), sound-isolating headphones/earphones are a better solution, allowing you to retain dynamic variations without much hearing damage risk compared to a headphone preamp to make it ultra-loud. If that's not suitable, you could introduce dynamic compression to lift the quiet passages in loudness, e.g. using foo_dsp_vlevel in the converter chain when creating files for your DAP (preferably from lossless sources, but possibly by transcoding from lossy to lossy, or from lossy to lossless or near-lossless (such as lossyFLAC or Wavpack lossy))

Here are a variety of approaches to using foobar2000 to pre-process your files.

1. Audiophile / obsessive. For zero chance of harmonic distortion. To retain dynamic variation and enjoy the emotional experience of highly dynamic music.
- set Album Gain with clipping prevention and use no DSPs at all. For better guarantee of equal loudness, set RG Preamp to -6.0 dB or so, because clipping is even less likely, even with a wide selection of classical music.

2. Practical attentive listening / high importance on volume matching. Little audible distortion or dynamic compression.
- set Album Gain without clipping prevention but preferably use Advanced Limiter DSP or simply accept rare hard cliipping that's usually so brief that it's inaudible. Could set RG Preamp to negative value to reduce incidence of limiting or clipping.

3. Louder attentive listening / accept some distortion.
- As with 2 but set RG Preamp to 3-5 dB but no more.

4. Very loud fairly dynamic listening / accept a little more distortion.
- As with 2, but RG Preamp 6-8 dB and must use Advanced Limiter to provide soft distortion limiting rather than risk hard clipping. Could use limiter prior to lossy encoder

5. Very quiet listening / noisy environment / accept dynamic compression to bring quiet passages up in volume but no need to aim for full-scale peaks and deaden the transients too much.
- Use foo_dsp_vlevel dynamics compressor with default settings or slightly lower strength. You could create dynamically compressed files in lossless format then scan them for ReplayGain and apply Track or Album Gain.

6. As loud as possible without a multichannel compressor / acceptable for weedy maximum DAP output volume.
- As 5, but crank up the strength and max multiplier in Vlevel settings. Could output to lossless then apply ReplayGain with positive high preamp and use Advanced Limiter.

I thought I wouldn't bother posting again after certain people merely skimmed my post rather than reading (which is fine on its own, but not when you reply based on that skimming!) however, I can see you took the question more seriously.


The issue is not formats. Very very few DAPs support ReplayGain. My Cowon D2 player is in the long list of those that do not (in fact, it would be simpler to list those that do as I'd imagine this list would be in the single digits...)


I absolutely do not want this. I am against the loudness wars in fact, yet this is exactly the sort of thing they are trying to do.


I just want peaks. I don't want algorithms trying to decide for me how my ears determine loudness. I find these algorithms to invariably be quite wrong. I find that peak normalization suits me best. Besides and more importantly, my player can barely drive my IEMs at all when some of the songs normalized to a too low volume come on and I have to adjust it close to the max (which has bad results when a properly normalized song comes on.) MP3Gain is fixing this problem for the most part though (though I had to find and enable the "maximization" option before it would stop trying to make them all have the same theretically perceived loudness.) I'm aware that I can reencode (or really even just directly edit) FLAC files, but my main issue lies in lossy files. Specifically, OGG Vorbis. So far as I've found, there is no MP3Gain for OGG files and reencoding is out unless I want to reencode to FLAC or something and just let them take up a lot of extra space. (Ultimately I may have to go with this, but my car has my old iAudio U3 which only has 2GB of storage, so that leaves a volume inconcistency there where I already have enough sound problems as it is...)

Anyway, foosions method seems to work fine for the PC. So at least I have consistency there where I can most enjoy it (which is good. I got the Promedia 2.1 speakers per some suggestions a while ago and they definitely are stronger than I'm used to. The last thing I need is to have it turned way up and have another song suddenly start blasting away at far too high of a volume...)


Well, it kind of will for the simple reason that it doesn't scan the entire song and limit to that song's peak. It won't compress heavily, but if it starts limiting at any point other than the peak, it has compressed downwards (and the fact that it is raised enough to cause this to happen compresses upwards) thus making for at least a slight dynamic compression.


Won't I then get clipping though? I mean, internally it's going to be 16-bits isn't it?



Not really. Please read rather than skim if you wish to respond to my posts. If you don't want to read them, that's ok. I understand that I write excessive amounts.


As I was saying, essentially it would be applying compression by its very nature. Let me explain what I mean since it wasn't clear. Compression in this case refers to essentially raising lower volume level points and lowering higher volume level points so that the entire waveform is essentially closer to the average level. Now, I was responding to foosion's suggestion of setting the ReplayGain to +12dB and then adding a limiter to the DSP settings. IF this were simply applied directly, it would be adding 12dB to the sound and then limiting it. Thus I had to ask because if this were too direct this would mean the whole sound was raised and then peaks shrunk. If you don't call this compression, what do you call it? Yes it's not quite as dynamic as mastering often involves as it would be far more constant, but you can't really call it anything other than compression under the circumstances. I think you can see why I did a double take at the suggestion?


It's because some are louder than others. Some have a peak of some 89dB or so, some have a peak of some 79dB or so. BTW, since it isn't clear, I do know how to read a VU meter and I'm no stranger to the likes of SoundForge, Goldwave, and etc. Once I realized that I needed the maximizing option turned on in ReplayGain, my problem with MP3s went away.


Yeah, COMPLETELY misunderstood that. What do you call someone who does the ripping of a disc? I call them a ripper when I'm being generic about it. I really don't want to call them "person who does the ripping of the disc." The problem is, people make mistakes. Or worse, they use crappy programs. (And, btw, copying CDs is not as accurate as you think. They have far worse error correction compared to data discs. This is how they can manage to get some 800+ MB of raw audio data onto a disc that in data mode would only fit some 700MB or so. The catch is, getting an exact copy is very very hard and few rippers know of the proper tools to do this even though the likes of EAC have gotten so easy to operate.) I suspect the biggest culprits come down to iTunes users and such though. I refer to the ripping program as "ripping program" btw, not "ripper."

So, you're stuck with a lot of lossy files and no lossless original because you didn't rip the files yourself. That is a different situation to many of us here, where we don't have to worry about transcoding because we can go back to the lossless source and apply DSP before encoding to MP3.

You also want loud playback on your DAP which can barely provide enough volume in your IEMs for your tastes, so even using Rockbox (supported on some other Cowon models, which provides ReplayGain tag support) wouldn't work. A headphone pre-amp would work, but you don't have one.

You say peak normalisation works for you better than ReplayGain. You probably accept uneven loudness to get the overall loudness higher and beat the background noise. Replaygain certainly works better than peak normalisation for creating even perceived loudness from track to track or album to album without having to reach for the volume control too often, but if the environment is noisy enough you might fail to perceive the loudness being even or you may need to raise the average levels (dynamic compression is the only option left) for some passages in the music.

If you're lacking in volume with your DAP I would really suggest subtle dynamic range compression (DRC), which is much less brutal than modern mastering practices. The foo_dsp_vlevel DSP at a low strength setting might be good.

Clipping of about 1-3 samples duration (at 44.1kHz) is probably going to be inaudible in the vast majority of cases, which is why I suggest using RG with 3-5 dB of preamp without clipping prevention. Encoder testing is done without clipping prevention and many LAME -V2 and Ogg Vorbis -q5 decode with clipping (because peak value is often about 1.2 x full-scale for most modern source CDs) with no loss of transparency. Otherwise these cases would be deemed "problem samples".

I don't know of any audible problem samples caused by clipping of the decoded MP3. They all seem to relate to characteristics of the original CD audio such as short transients and not to the fact that clipping occurs when decoding and restricting to 16-bit. Only excessive additional gain provided by setting an mp3gain target (or fb2k preamp) very high would cause audible problems.

To achieve the DAP loudness you desire, you have to accept some compromise. Allowing SOME clipping would probably have little to no detrimental effect on the music, which is why I suggested it in option 3. It also doesn't require a transcode if your source is MP3 or AAC-LC (common .mp4 or .m4a encodings) because mp3gain, aacgain or foobar2000 can apply global gain field adjustments to increase the loudness without transcoding. (Vorbis works differently, so vorbisgain is only tag-based RG which the player must support, or you must encode from a RG-adjusted source, such as with foobar's converter)

I'd say the options in my previous post are in order of decreasing sound quality / increasing alteration from the source audio, but even numbers 5 and 6 can give an enjoyable audio experience that's better than struggling to hear the sound and better than modern CD mastering practice if you have a suitably dynamic source. Even transcoding can allow a pleasant experience. I'd suggest a double filename extension to indicate the original source format and that it's transcoded.

Transcoding in rough order of best to worst: - based on others' experiences and investigation, not my own ABX tests.

NO TIME-SPREADING, NO FURTHER LOSS:
1. Transcode to lossless, e.g. FLAC or ALAC as supported by your DAP. No worse than listening directly to foobar output, but bitrate is about 700kbps or more (less if you use modest ReplayGain targets than if you use high volume).
NO TIME-SPREADING, only added noise via near-lossless encoding
2. Use lossyWAV/lossyFLAC/lossyWV. Transcode using lossyWAV followed by a compatible lossless compressor like FLAC (if your Cowon D2 supports FLAC). Probably audibly indistinguishable at --standard setting or above (typ ~460 kbps). In noisy environments (e.g. DAP on public transport), the --portable setting (~380 kbps) or even settings as agressive as -q 0 (~290 kbps) may be indistinguishable from uncompressed, though Wavpack hybrid (see below) might be better, if it's supported. Should avoid conventional transcoding artifacts at a bitrate close to high-bitrate conventional lossy.
3. Transcode to Wavpack hybrid. Supported by only a few DAPs but all with Rockbox, and doesn't measure the noise floor like lossyWav --standard. Probably hides the added noise better than lossyWav --portable or -q 0 so could be better at same bitrate. Can get pretty good quality as low as about 250 kbps.

TIME SPREADING and other transcoding artifacts/unmasking possible
4. Transcode to different conventional lossy format at high bitrate. Choose another supported lossy format with a good high-quality setting, e.g. VBR quality setting that normally gives transparency or near-transparency. Even going from MP3 source, many MP3-only players can actually handle MP2 files, sometimes only if renamed as .MP3, so it might even be an option for the most basic DAPs and digital radios with MP3 support, though MP2 is poor below 192 kbps and has inefficient dual stereo mode for quality with no safe joint stereo mode.

5. Transcode to different lossy format at moderate bitrate. If bitrate is quite important to you, choose a lower setting from a good encoder, e.g. LAME -V5 or AAC 128 kbps.

6. Transcode to the same lossy format at high bitrate. Using the same format seems to cause problems (e.g. mp3 to mp3 and aac to aac are both worse than mp3 to aac or aac to mp3), but sometimes only one format is acceptable for compatibility reasons. Padding with a few milliseconds of silence can mitigate some but not all of the adverse effects, apparently. Encoding at extreme settings such as LAME -V1 or -V0 (beyond the standard just-transparent setting) can help reduce the potential unmasking artifacts a little at the expense of higher bitrate (e.g. 200 kbps or more).

7. Transcode to same lossy format at 'standard' setting. Using LAME -V2 for example, or OggVorbis -q5 is considered transparent normally, but can be unmasked by double-encoding.

8. Transcode to same lossy format at low bitrate. Using LAME -V5 for example (~130 kbps) you might get fairly good results or some nasty transcoding artifacts from time to time.

My first recommendation:
Don't get too hung up on clipping-prevention-at-all-costs. With modest RG pre-amps (or mp3gain set to as high as 92 or 94 dB the clipping isn't likely to be audible, especially during the dramatic peaks where it's most likely to occur.

My second recommendation:
Don't be too afraid of dynamic compression. A little subtle DRC can improve the listening experience in noisy environments of with weedy DAP volume without seriously degrading the overall emotional journey in the music or robbing it of its punch, kick or sparkle. Check to see if your DAP incorporates any DRC options.

My third recommendation:
Don't be too afraid of transcoding. It may still be the a part of the most practical way of maximising listening pleasure within the constraints of certain situations.

My final recommendation:
Don't be afraid of strong dynamic range compression. In difficult acoustic environments where volume must be restricted or background noise is high, or you must prevent hearing damage, it can genuinely provide the best sound quality possible in the circumstances. Even coupled with transcoding, it might be the best solution possible in such circumstances and provide a pleasant sound without causing listener fatigue.

If you're really stuck with a DAP that's too quiet for you, and listening in a high-background-noise environment, I'd imagine very dynamic tracks (like Queen's well-known Bohemian Rhapsodie) would end up lost in the noise during the quiet parts. Then, I'd truly suggest foo_dsp_vlevel is worth a try (on a subtle setting) followed by the least harmful transcoding setting that you can manage (both from point-of-view of format support on your DAP and from bitrate that you can accept, given the quantity of files that you have to transcode versus those you can leave alone, which might allow 700 kbps for a small number of transcoded files, while the rest (e.g. modern compressed recordings) remain at 128 kbps, say, in their original format) with, for example mp3gain and a little clipping permitted.

Ideally, you'd quarantine the files created for your DAP in a folder called, for example "C:\Cowon D2 Music - clipping, DRC & transcoding possible" so you don't play them on your PC when you have the superior originals instead. You could also use informative file names or double file name extensions, such as:
01 - RockorPopTitle - Artist1.RG94.mp3
09 - DynamicTitle2 - Artist2.ogg.vlevel10%.transcode.mp3
04 - DynamicTitle12 - Artist3.mp3.vlevel10%.lossy.flac

Perhaps you'd create the second of those two then apply mp3gain to it to ensure the volume is matched to the other files you own to which you didn't apply vlevel.

With the third, to match 94 dB ReplayGain, you might have first applied vlevel while outputting to lossless, then recalculated ReplayGain and applied it when converting to lossyFLAC by way of lossywav with stdin input, its output piped to FLAC, after which you'd delete the intermediate lossless FLAC which had only had vlevel but not RG applied.

An optimal approach would probably be:
Step 1.
Keep originals safe. Copy all files to the DAP folder on your PC.

Step 2.
Apply a high ReplayGain like 94 dB without clipping prevention to the mp3 and aac (.m4a) files themselves using foobar2000, mp3gain or aacgain. (OK, for the Cowon D2, I believe that AAC isn't compatible, so you'd be forced to decode to FLAC or APE or transcode to lossyFLAC, mp3 or ogg vorbis)

Step 3.
Apply ReplayGain while converting any lossless originals you have to the destination lossy format (mp3 or ogg vorbis would make sense for the Cowon D2, as might lossyFLAC if FLAC's simple decoding is found to extend battery life).

Step 4.
If you have Ogg Vorbis source tracks/albums that show a ReplayGain of -3 to -7 dB, they're definitely close enough to leave alone given that the D2 doesn't support Vorbisgain (I'll take your word for this). In fact, if aiming for 94 dB SPL, I'd be tempted to leave alone vorbis albums with Album Gain values of -2 to -8 dB. Otherwise, if you need to adjust the volume of Vorbis sourced files, you'll be forced to convert to a large lossless file (FLAC/APE) while applying RG in fb2k (which will sound as good as the vorbis), or convert to a smaller lossyFLAC or transcode to mp3 (e.g. LAME -V1 or -V2 should sound pretty good if FLAC or lossyFLAC isn't acceptable).

Step 5. If you find any files that generally seem loud enough but where clipping is audible and annoying at the gain you require (hopefully you won't, though), try going back to the source file then in fb2k's converter apply Replaygain and try applying Advanced Limiter to soften the clipping distortion, then convert losslessly (FLAC/APE), or to lossyFLAC or transcode to a different lossy format.

Step 6. If you find files or albums that are too dynamic for the listening environment required, go back to the source files for those and use fb2k's converter with foo_dsp_vlevel at a subtle setting (Configure selected DSP) and convert to lossless. Then scan that lossless output track/album for ReplayGain. Then convert that lossless file with RG applied into the desired output format (lossless FLAC/APE, lossyFLAC or transcode to mp3 or ogg - whichever is different from the original source format).

Hopefully you'll get away with only Step 1 and Step 2, and perhaps Step 3 and won't need to adjust the volume of any Ogg Vorbis source files. That way you avoid transcoding and any form of audible dynamic range compression. The other steps, if necessary at all, will probably only be needed for a small proportion of your music unless your environment is so noisy that it's worth using vlevel on all your music, so the extra effort in reprocessing those files won't be too arduous.

But really, don't worry if you need to apply DRC for noisy environments. The music can remain enjoyable even when slightly compressed and transcoded, and will surely be far more enjoyable than music you can't hear below the volume of the background noise!

That should pretty much cover all bases.

One more thing from this review of the D2 is that it seems that if you can restore the factory settings and select USA instead of a European country you might be given a greater volume output to help overcome the limitations of your IEMs in combination with the D2. (possibly thanks to an impedance mismatch)

My D2 can provide more than enough volume for my IEMs -- I often use about 38 or so out of 50 on the volume. My D2 CAN'T perform miracles and make a song that is really really low volume to begin with be where it should be. A song normalized to some 89dB or so at its peak will be quite loud and the D2 has no problems whatsoever. One normalized to something closer to, say, 69dB will sound really quiet and I have to turn the volume way up. As you say, I don't have the originals for some of these anymore (and one that I do have I haven't been able to rip without so many errors it drives me crazy due to bad scratching) so I can't normalize without going from lossy to lossless since I refuse to go from lossy to lossy (some are from the earlier days where the few services willing to offer anything without DRM insisted on 128Kbps because some jerk arbitrarily declared that this is "CD quality" a long time ago.) I found that with the replaygain algorithms especially this was a problem because even the peaks would be lowered so far that I'd have to really really turn it up to get a normal volume level. (And, btw, the D2 does have one of the best output levels of any portable player, but it's not even on the same scale as my PC's soundcard or the headphone output of my new speakers each of which probably could blow my IEMs if I were stupid enough to turn things up as high as they could possibly go. These can compensate for some pretty bad normalization, but the D2 can't compensate for the worst of it.)

BTW, I do have a preamp (though it's a crappy one I mostly just use for things like my DS and EeePC with their really low output levels and I'd have to get a better one if I were listening this way often) but the thing is that kind of defeats the purpose. My IEMs and DAP I carry around in a pocket all of the time. Adding a headphone amp -- even a small one -- would be something of a problem.


What background noise? I'm using IEMs... No, I just want peaks to be at the same places. If one is more dynamic than another so is overall quieter, that's fine. I don't want equal "loudness" I want equal peaks. To my ears this just works better and I can always adjust the volume controls manually if I want more loudness.

Replaygain certainly works better than peak normalisation for creating even perceived loudness from track to track or album to album without having to reach for the volume control too often, but if the environment is noisy enough you might fail to perceive the loudness being even or you may need to raise the average levels (dynamic compression is the only option left) for some passages in the music.


Vast majority maybe, but at times it can get really bad. Clipping really bugs me so even just hearing it rarely is undesirable for me. Yes, it may be inaudable most of the time, but it only requires some of the time to be annoying. I'd rather lower volume levels than clipping really.

OK, so volume isn't the issue. You actually don't think Replaygain works. I'm at something of a loss, as I think it does.

It chooses a point at about 95% of the way along the list of loudness values from quietest to loudness, so represents something close to the highest 'instantaneous loudness' and is usually perceived to be pretty darned consistent at matching a human impression of loudness.

Digital peak normalisation is very dependent on the precise details of the shape of the highest peak in the file and bears little relation to the overall loudness unless you listen only to heavily dynamically compressed and heavily limited music with a tiny peak-to-average ratio where it will work by coincidence alone.

We clearly haven't been put in the picture about your aims in a way we understand, so most of the advice you've had in this thread is about increasing loudness on a low output DAP when that's not what you want at all.

If you want more even loudness from ReplayGain, I'd suggest turning down the pre-amp (negative values, aiming for lower target volume to prevent peak values from reducing the gain if you have highly dynamic music. 83 dB SPL was the original target (now it's 89 dB SPL to be closer to loud music) so a pre-amp of -6.0 dB would be good to accommodate highly dynamic classical music without peak clipping.

What you mean by peak and what Replay Gain means by peak are two different things.

You mean the highest loudness you perceive during the track or album. You can't possibly perceive the sample value of a single sample peak, which is what it referred to by Replay Gain's peak value (simply there to help prevent clipping).

You're in fact talking of peaks at 89 dB and the like, which implies you're referring to instantaneous peak perceived loudness. Replay gain approximates perceived loudness as the average frequency-weighted sound energy over about a 50 millisecond window. It doesn't report the peak value of the instantaneous loudness, so you can't use it to normalise your files for maximum instantaneous loudness being the same. ReplayGain throws away the array of loudness values once it has sorted the array and chosen the 95th percentile value.

Your idea sounds more like an analogue peak meter - fast responding, but still averaging over at least hundreds of samples in a rather perceptually relevant way, not like a digital peak meter which can show and hold single-sample peaks that you might never hear. I don't think there's a commonly-available tool to automatically the peak perceived loudness

Your approach on a rhythmic track might lead to the loudest drum hit determining the vocal level. For example The Rolling Stones' Mick Jagger screaming "I need a love to keep me Happy" on a dynamic 80's/90's CD release or original vinyl would be desperately quiet compared to the almost whispering Dido's breathy "I-I-I want to Thank You, for giving me the best day of my life" because the latter is compressed to squash the drums to rather dull thuds and raise the vocals, while the former is only slightly peak-limited to allow seriously sharp percussive drum hit that is briefly MUCH louder than the clearly top-of-the-voice vocals.

Replay Gain would get the vocals pretty much at the same level and let the Stones' drums ride higher, which to me is the right approach.

Interestingly, if you used peak digital sample values to normalise the same two tracks, Dido's drum hits would still be much louder, because the peaks on the record are heavily limited and flattened/widened so contain more energy when scaled up to full-scale than the Stones' sharper narrower/taller peaks (which are only slightly limited, mostly with soft tape saturation). When scaled to full-scale height, the narrowness will mean they contain a lot less energy than Dido's drums and will be a lot quieter. The Stone's vocals will be quieter still.

Anyhow, I don't think there's a tool to do what you want unless you edit the source code for a Replay Gain implementation and use the 100th percentile (simply the loudest instantaneous loudness measure).

Most people are happy with Replay Gain so there's little incentive to implement what I presume you're after.

*sigh* And we start with this again. Ok, look, there's not really much that needs to be discussed here so I'll make it short so no skimming is needed this time. The only problem I have is with the DAP since it lacks any software correction methods. And no, I'm not worried about noisy environments (ok, the car is pretty bad, but mostly I'm just aiming for consistent volume levels there since the line input is pretty weak and songs normalized too low require me to turn the volume all the way up to the maximum.) MP3Gain mostly solved the issue with MP3s (though I found it a bit annoying that it has to work in such large increments -- namely it bugged me because for some reason it would once in a blue moon allow a sample to clip due to rounding up to the nearest increment. But I never noticed it, so I assume it's just one or two samples here and there at the worst.) I was unaware of the VorbisGain utility (which is why I asked earlier if there was an equivalent for it) so I'll give that a shot. If it doesn't work, I'll just give up and use FLAC despite the low storage availability on the U3. It's only a few that were a problem anyway. EDIT: Yeah, it seems VorbisGain only cares about equal loudness too. I couldn't find a "maximizing" option. I tried setting it extremely low or extremely loud just to see what would happen and nothing happened.

Just to be 100% clear. I ONLY want peak normalization. I do not want loudness normalization. Period. That's it. I was merely hoping that ReplayGain would give me a nice software based solution on the PC/laptop without having to modify any files. That's all. I never said it didn't work or anything like that, just that it doesn't suit my needs when using the equal loudness method.

BTW, it does seem as if my version of Foobar2000 is ignoring the replaygain preamp settings.

The reason people are just skimming through your posts is because you fail to accept the answers provided, and only rely on your own homemade theories. You can get all the help you want here, but as I mentioned in my first post, you will have to accept that you may be wrong.


I could say the same thing about you when it comes to skimming. I've had uneven volumes myself back when I used MP3Gain with odd values and foobar2000.

The reason that you THINK you have a more even perception when you use Peak normalization, will only apply to any new music mastered with high compression.

Really - yes you DO need replaygain if you want to make sure that all your music is the same perceived level. You may have experienced odd values because of MP3Gain. Eventually try to re-calculate the replaygain on all your files with just foobar2000.

In reply to your question what I would call the process of gaining and limiting - I would call it limiting - Not dynamics compression. And this will produce a very bad output. At least, if you REALLY want to compress further more (or put some compression on older material that are not as compressed) you should use a *real* Dynamic compressor like WAVES VST (using VST wrapper).

In reply to your ripping trolling, I'm fully aware of the techniques of ripping, still I fail to see the connection between dynamic compression and ripping (and now you bring secure ripping into the discussion? Please stick only to the subject, if you want useful replys).


My recommendation is that you control replaygain on all your files with foobar2000 only (eventually only using MP3Gain to slightly adjust only to prevent clipping), then when you need files on your DAP, use foobar2000 converter with a Dynamics compressor (as mentioned above) to ensure that all files, old and new music will all have the same percieved volume as loud as possible to make your DAP play it at a decent level, yet on the cost of a little lost dynamics (new music with high compression, would not be changed much when another compression is applied).

Sounds like you forgot to set "Source mode" and "Processing" in replaygain settings - Also did you make sure that your files are replaygained by foobar2000? There happen to be differences in the standards used to store the RG value (some use ID3v2 other use APEv2)

As you said, your DAP lacks software correction methods. Therefore it ignores the vorbisgain value which is ONLY stored in a tag. You can't gain-adjust Vorbis files without re-encoding, which puts it at a disadvantage compared to MP3 and AAC-LC, so in that sense vorbisgain is not equivalent to mp3gain. You only option is to leave alone or re-encode (i.e. transcode or store lossless or near-lossless).



The increment of 1.505 dB is inherent in the mp3 standard and is the useful feature that enables mp3gain to adjust safely without re-encoding (unlike vorbis). It means you're accurate to the nearest 0.753 dB which is good enough for loudness adjustment on a human scale.

Sometimes the original mp3s clip anyway even without adjustment thanks to the lossy nature of the mp3 format, so it's not always mp3gain that causes clipping, as music measured at 89 dB and therefore unadjusted could still clip.




Well on the PC/laptop foobar will do just fine for peak normalisation (in fact peak-maximisation) almost all the time so long as you apply Clipping Prevention (on the track or album basis as you deem appropriate) and set the largest possible RG preamp. It can also help you adjust mp3s for use on unsupporting DAPs, whereby the peak will be adjusted within 0.753 dB or 1.505 dB of full scale (I'm not sure which, and don't really care). Foobar cannot adjust Ogg Vorbis files for your DAP, so you will quite simply have to transcode or decode to lossless or near-lossless if you need those files peak-adjusted for your DAP.

Many of those rare highly dynamic and barely peak-limited tracks with very low volume that you can barely hear on your DAP may still have full-scale peaks so these will no get increased at all by peak normalisation, which is exactly how digital peak normalisation ought to work (i.e. completely ignoring how loud the sound is, just setting the highest sample value to maximum by linear scaling of the whole track or whole album). Then you'd need to either apply gain manually and accept clipping of the rogue peak (use mp3gain's apply gain function) or in foobar, turn off clipping prevention and use advanced limiter, or use a dynamic compressor, or isolate and reduce only the one enormous peak by adjusting only the gain of that frame using a util such as mp3DirectCut. Then the RG peak will be reduced and you can scale up the loudness without clipping anything else, while only affecting the 26 ms frame in which the rogue peak occurs.

I think the rest of us should ignore your topic title, because ReplayGain's volume normalisation and whether it works or not is clearly not what you're interested in. That has probably led us into discussions you don't wish to get into but that we thought you did.

From my point of view, it's topic-closed. I think you have all the tools to do what you want to do, except for what can't be done (i.e. scaling vorbis files on your DAP without transcoding). If you apply this and don't get the loudness you require or expect on certain files, then peak normalisation alone isn't the right approach for you on those files. Best of luck.