Can some one please tell me what the 'Use 6db Hard Limiter ' option is? I have absolutley no idea. Also, it would be a good idea if we had a 'sticky' telling every one what every option was. Thanks

Look towards the end of this thread.

"A hard limiter is a compressor. It reduces gain above a given threshold of loudness, limiting all peaks above the threshold to the threshold level. So the high volume parts would get limited, and the low volume parts can be louder. This way you can avoid clipping, and still get the maximum possible loudness for the song, a technique which is, in some variations, widely used in radio and TV broadcast. "

*YUK* - disabled

PS thanks john33

Yeah, better not to use it, it audibly changes the sound.

i see placebo BS science spreading.
hard limiter can audibly modify the sound indeed, but it completely eliminates any clipping. also, if your volume is set to -6dB or lower, hard limiter triggers only in places where samples would normally get truncated. conclusion ? set wave volume to max in windows volume control, and use fb2k's internal volume control instead, it eliminates clipping most of the time, and makes hard limiter never (or hardly ever) trigger if enabled.
maybe there are different points of view on this matter, but i find hard limiter less annoying than clipping or even auto-attenuation.

Woo! The man himself replied to one of my topics! Beat that To be completely honest I couldn't audibly tell when it was on but I just feel much more happy knowing it not doing something with my sound or wasting precious CPU cycles (P2 350Mhz ) Resampler takes up 50% CPU

I don't

Peter, could you make the limiter configurable?
It would be nice to be able to set it to 0.01 dB.
Such setting would remove clipping too and wouldn't make sound quieter.

Remember that dB is a logarithmic scale. EDIT: Never mind, you, being an audiophile, most likely know this.

Also, you shouldn't really be experiencing noticeable or possibly even any clipping if you ReplayGain scan your files. Depends on the average perceived loudness of the material, though, so it may still clip or hit the limiter.

I'm afraid it doesn't work that way.
0.01 dB is only 38 steps or 1.15% from full scale.
6 dB is about 16,000 steps or 50% from full scale.

Imagine a graph of input level on the x-axis against output level on the y-axis, as a percentage.

The y axis is restricted to -100% to +100%
The x axis can go a little beyond 100%

Ideally, you want to map input values (X) to output values (Y) on a 1:1 basis so the output sound is equally loud.

However, if the X values go over 100%, you get clipping as the Y values cannot exceed 100%. This means a smooth sweep (e.g. ramp or sinusoid) beyond 100% on the input turns into a smooth sweep that suddenly stops rising and becomes instantly flattened on the output.

This sudden change of slope introduces completely new frequency components into the sound and generally makes it sound awful. If it's only a couple of samples at the very tip of a wave, it might be perceptually inaudible.

I don't like to take the chance, so I use Replaygain (I'd use it anyway to make the perceived loudness more consistent between albums of different types and eras, but you can simply enable only the clipping prevention function).

Replaygain tags tell you the maximum sample value in a file and allows you to choose to scale the file slightly so it won't clip (so the 1:1 ratio of Y to X becomes, say 1:1.2 if the maximum excursion on X is known to be 1.2). This definitely introduces no spurious frequency components. A 20% change in scale is a barely audible -1.6 dB reduction of loudness. Because you know the maximum value of X (or minimum value, if it's a bigger excursion from zero), ReplayGain clipping prevention ensures you will never try to go beyond the full scale on Y by allowing the slope to change to keep it within Y = +/- 100%. Because 16-bit audio has a tiny noise floor, the change in level is negligible, and dithering can preserve the full audible dynamic range.

If you instead set up a hard limiter with a -6dB break point, it will allow distortion of the wave for large samples. So the straight line of 1:1 slope on our graph starts to distort. An X value of 30% (below the break point) will map to a Y value of 30%, but an X value of 70% might map to a Y value of 65%, an X value of 90% might map to a Y value of 80%, X of 110% might map to 95%, and so on. You can see this has curved. This causes squashing of large waveforms. It tends to deaden large transients and tends to add harmonics to signals that weren't in the original sound. If it's done right, some people like this - it sounds like an overdriven (gently read distorting) valve amplifier or audio tape. It adds warmth to the intended sound and isn't as nasty as hard clipping at 100%.

The sharper the transition from straight line to curve, the worse the distortion and the more it affects the sound. A 0.01 dB break point adds massive distortion - almost as much as a hard clipping (i.e. no hard limiter). That's why it's done at 6 dB instead. You can't get something for nothing, I'm afraid and the downside of 0.01 dB clipping would be massive.

Compressors and limiters can also rob some of the dynamic variation from the music where a quiet bit changes to a surprise louder bit with an emotional impact. That's what's happening with some of today's ever louder, more compresses radio mixes of popular music. Perhaps that's why music sales are struggling - the magic has disappeared in this arms race for ever higher loudness at the mastering stage. Loudness has a short term impact on perceived quality, but variation and surprise gives music its long-term power to maintain interest and excitement.

Dick Darlington

The proper answer is: we should use replaygain combined with limiter.
Limiter set high enough, so that it won't distort normal sound, eg. 3 dB.

<edit>
As for loudness... just use your amps volume control
</edit>

No, better answer is: simply setting preamp to -2 dB, you should avoid any possible clipping. No need to use hard limiter then.

RG is sufficient.

I assume the limiter code that FB2K uses is a variant of the limiter code I originally wrote for XMMS ReplayGain support, so a graph of the response is here:



As long as the original is not close to full scale, the slope is pretty smooth, so the limiter won't have any real effect until the music would be distorted without it, when it'll compress the music. I'm surprised some people (well, one I can see) feel that clipping sounds better than compression. For most people it'll be exactly the opposite.

So remember: the limiter kicks in when not having it kick in would be worse!

Sometimes limiters are used to compress music and make it sound louder, but that's not what the one in FB2K does!

In principle best solution is to use ReplayGain, and leave hard limiter enabled. Replaygain will lower volume enough that the limiter will only trigger when it's really absolutely necessary.

For a lot of contemporary music this is completely false.

Not according to my (non thorough, I admit) tests. If you set max. amplitude to -2 dB, you have a headroom of 2 dB for DSP increased max. levels, which, according to my tests, is enough for these cases of just removing data.

But maybe I'm wrong, could you give me an example where this could be not enough? I think it would be good to have this issue cleared up.

I don't think you are practically wrong, but I'm not only talking about DSP effects. The music itself can also (and very very often does) clip!

I don't know if we are understanding each other... About DSP effects, I mean that after applying DSP the max. level of the music can be increased, and pass the max allowed value, in this case the music will clip. Lowering the final max. volume 2 dB, you have a margin of 2 dB for this increased max. level after the DSP.

If you are talking about music that was clipped already before any DSP, well, in that case there's nothing that can be done to avoid this clipping.

Some more tests: I've done a quick test with a heavily clipped song in origin, compressed it to LAME --aps and --cbr 128, and using MP3Gain, have verified that a reduction of the output amplitude of -2 dB at decoding results in no clipping at all, by a quite safe margin.

I hope this clears up the issue.

Edit: the problem about using the hard limiter is that it will kick in 6 dB before the music will clip, and that's quite a lot. With a proper Replaygain, there will be no clipping unless you apply some additional DSP in the player, so the hard limiter is not necessary. Without Replaygain, the hard limiter will be distorting continuously any modern cd track, even if it's not clipping. And it's quite audible, believe me.

Well, yes, you can apply the limiter to trade compression with clipping. That's the point of it.

But that music is supposed to clip, so if you do that, you are in fact changing the music. Also, the clipping effect will be reduced but will still be there, unless it clips just at the end of the limiter curve.

One song is not representative at all. I have a very popular Red Hot Chili Peppers CD here that will clip even with 12dB reduction. Average case is no problem - it's for the bad stuff that the limiter is a safeguard.



6dB is not quite a lot. The majority of consumer soundcards have internal clipping way before full scale level (source: Frank Klemm). Even with ReplayGain, there can be clipping if the track is very dynamic (but this depends on how the player is programmed - you can automatically reduce replaygain to prevent clipping, WinAmp had this option, but then you no longer have guaranteed equal loudness).

I don't argue that the hard limiter modifies the sound of the music - but without it and without ReplayGain, 'any modern cd track' will have heavy clipping. With it it will be compressed a bit. IMHO compression is much preferable to clipping, although apparently that's a matter of preference. I could imagine that if you are used to contuniously listen to clipping music that the compressor will sound different and bad.

No, this is completely wrong. The PCM data on the CD does not go 12dB over the maximum level of the CD! (it simply can't go over the max level, period). The heavy clipping is the result of the lossy psychoacoustic compression/decompression combined with already heavy compression of the original signal.

Clipping effect does not completely disappear with the compressor, but most of the bad distortion is smoothened out.

PS. I'm interested in a sample where clipping sounds superior to limiting to you, or where limiting causes bad degradation of the original track.

Wow! Could you post the piece of it where the clipping goes that high?



!! A card that clips at -1dB nowadays is crap, it will sound horrible with many modern music you play. I can play a signal at 0dbfs with my Audiophile without any clipping, and harmonic distortion below 0.01%.



I don't think *any* will have clipping if you reduce the level a couple of dB at the decoding stage. Maybe some, and I'd like to see one example, just for curiosity.



I don't usually listen to clipped music, because I usually listen to wav or cd. And if that music is clipped, it's because the mastering engineer or the artist wanted so. That music is supposed to clip, if you avoid this, limiting the signal, you trade planned distortion (clipping) for added distortion (limiting) not present in the original music.


That's a different case that I was talking about. In your case, the additional clipping is caused for the additional DSP (compression).



Play a 1 KHz sinewave at -1 dB. You will hear it sound clearly different with and without hard limiting. Try any modern song that goes up to nearly 0 dB, uncompressed. It will sound different, with some added colouring soft distortion. The only way that the hard limiter doesn't make this, is to reduce the level of the sound at least 6 dB at the decoding stage. This case, replaygained songs that can still clip, is the only where a limiter could be useful, but I'd like to see how common is this. A few clipped samples in a noisy or loud passage are really difficult to hear. And for listening uncompresssed audio, the hard limiter simply degrades the quality of the signal.

I'd like to see how many songs have clipping just decreasing 2 dB the output at the decoding stage. Maybe some, but that's what I'd like to know, how many, how audible would be that clipping, and what reduction would be needed to avoid this.

Uh, that doesn't look trivial - I'd need to cut it out the mp3 since decoding to wav will just clip to full scale audio, or I'd have to retrieve the original CD and take a guess at the settings used to encode.

(12dB is exceptional though - have only seen that a few times)

A random pick from my samples directory gives me Eels - Souljacker which decodes to +2.75dB over full scale. Not close to 12dB, but your 2dB headroom is already too small. The clipping is clearly audible to me.

The 12dB track is 'Red Hot Chilli Peppers - By the Way - Venice Queen' BTW.

REPLAYGAIN_TRACK_GAIN = -9.7400 dB
REPLAYGAIN_TRACK_PEAK = 4.5812



With an Audiophile, maybe, but that's unfortunately not a common consumer soundcard



What's 'a couple'? You have to make a bet to a safe margin. The next year the CD's are louder again and boom there goes your margin. Moreover, if you reduce it too much, you'll have problems with people that don't have powerfull enough amplifiers. Been there, done that (with ReplayGain 83 vs 89dB issue). I don't like this kind of solutions.

If you can afford to, then yes, reducing level is always a good solution. If you don't like the effect of the limiter then it is better, as long as you take into account a bad track can go over your safety margin.

Note that if you reduce the dB, the limiter will not have any more effect on the music! So saying 'it's bad' makes little sense as it disables itself in your solution.



Ehm, this is factually wrong. If the music is preclipped during mastering, the limiter will preserve that clipping, although it will not be at the full scale of your output. But the planned distortion will not be affected! Feeding a square wave through a limiter gives you a square wave back (but at a lower level).



I don't understand at all then. What *are* you talking about?



Music please.



Aha, maybe this is your point? For uncompressed data without any DSP in use, it makes no sense to use the limiter for the obvious reasons that there can never be unintentional clipping.

For compressed music I would recommend enabling the Limiter with or without ReplayGain.

PS. Reason why I'm especially annoyed at this thread and you is the following:

Person comes in and asks what limiter is for.

Gets helped by a quote that doesn't relate to what limiter in FB2K does. Then you comment 'yeah don't use it it changes the sound'.

That's about as braindead as telling someone 'mp3pro sucks don't use it' because it's unfit for your high bitrate archiving needs. What if he has a 32M portable? Oops.

Limiter is there and enabled by default because for vast majority of users (not replaygained contemporary music MP3's) it provides better and not worse quality sound.

BTW. You can sort on track peak if you have replaygain info in your database.

My winner:

Gus Gus - This is Normal: REPLAYGAIN_TRACK_PEAK = 7211.5311

(admittedly probably a corrupted file )

Top 3.5:

Beethoven - Moonlight Sonata: REPLAYGAIN_TRACK_PEAK = 5.3847
Bush - Machinehead: REPLAYGAIN_TRACK_PEAK = 4.9135
Beethove - Pathetique: REPLAYGAIN_TRACK_PEAK = 4.7552

I have over 100 that have >3dB clipping.

Set preamp to -6 dB to avoid any compression with normal music
or replaygain your files and set Preamp to -2 dB to give headroom for those loud/album files.
Don't forget to turn the limiter on as a precaution.

Ok, I've done a quick check of some songs, and I don't have time to write more right now, sorry. I've checked with mp3gain around 30 mp3 songs, and found a few that go over +2 dB, one in particular (Music from Madonna) that supposedly goes up to +4.5 dB according to mp3gain. Strange thing is, I have replaygained this song -2 dB, and whilst mp3gain says it's still clipping, I've decoded it to 16 bit with LAME, and found no clipping at all!! Even checked it with CEP statistical analysis that tells even if there will be clipping between two samples (after analog reconstruction), but nothing, by a pretty wide margin.

With -6dB and no ReplayGain, you will still have (some very slight) compression, though only on the peaks where you would normally have clipping.

Might be interesting to note that I checked my version of "Venice Queen":

+1.2 dB according to MADPLAY.

I also checked my "Music" version:

-0.3 dB according to MADPLAY.

reminder: you don't even need to use preamp DSP, volume control is applied *before* limiter (unless you enable "directsound volume control" in config which is off by default)

Eek! I've done my best to fumble through everyones replies but most of it is beyond my knowledge. All I wanted to know was if the limiter would change my sound at all. I have a pretty hi-end setup; good amp and brilliant speakers [my pride and joy] and was just looking for someones advice on what to do. Yes, I do have my own ears but I was just looking for what other people thought / suggested I do to get the best sound. I'm not bothered about volume [I'm not that lazy - I can lean over and turn up my amp]

unless you're 100% sure that you will never ever have a song clip you are better off leaving it enabled. It barely makes an audible difference, if ever, but it will protect at the very least your speakers. The only way to be sure you will never have a song clip is to go through every single one and see what the absolute peak of all of them is and adjust the volume control in foobar as necessary to avoid that clipping. It's a much simpler solution to simply leave the limiter on. I've got a bunch of songs that need +1 or 2dB according to Replaygain, and they already have a peak of 0.9xxxdB anyways, so with Replaygain they'll peak at almost 3dB above full-scale. That is unless I have the limiter turned on, which I do. They don't clip because of the limiter, and they still sound fine to me.

I don't use replay gain and [correct me if I'm wrong] I would never have my amp up high enough for clipping to harm my speakers. Perhaps I've greatly misunderstood the whole thing though...

Correct me if I am wrong, but why not use the preamplifier component to prevent digital clipping?

Set preamp to -6 dB.
Leave the limiter on to be extra safe, it will probably never activate.
Crank your amp up 6 dB.

A negative effect of this is that (on a 16 bit card) you lose one bit of precision, i.e., take these 6 dB right out of dynamic range and S/N. But if your brilliant speakers are fed by a brilliant 24 bit card, this is not an issue.

-- Oops, should read previous posts.

'fraid not mate, it's a 16 bitter And my speakers are brilliant, Elxat Synphony 6.2 Mmm... Lol

when clipping's involved, it doesn't take much to fry tweeters, clipping should be avoided whenever possible

Yes, this sums things up pretty well.

I've sent Peter some of the badly clipping files and asked to check whether it's not a decoder problem or corrupted files which makes it hiccup. If it is, KikeG's suggestion of just setting preamp to -3dB and disabling the limiter may also be fine in practise. But the above is definetely the way to go if you have a good amp-soundcard combination.

Greetings.

I just wanted to note that if I see further discussion of this sort in other threads, please let me know so I can split them off into another topic, and avoid this kind of annoyance for all the users.

I don't always see every thread, but that is what the "Report to a moderator" button is for. If you think a thread is getting out of hand or off-topic just let me know.

Sorry for making such a comment. I was so naive as to think this forums were for discussing things Mmm... I though we were talking about hard limiter, and the discussion derived into pros and cons of it. I like to discuss such things, I get sometimes quite involved, but where did I go off-topic??

Edit: sorry If I was too ironic in the first sencence, but I truly believe my post was not that inadequate.

I don't think the discussion was offtopic at all. My complaint was about the way there had been replied to the original question.

OK, I guess my response should have been more detailed then. I had talked about this same thing previously with more detail, and I was in a hurry, so I thought this response was enough.

So, I've been doing some more testing over this, and here is what I've found.

Ok, I must admit that the figure I proposed of -2dB of headroom is not enough in case of highly compressed audio, because here it can lead to too much clipping in some cases. In case of high-bitrate compressed music, it's probably enough to avoid audible clipping in most cases. Anyway, I think -3 dB is a better figure for all cases.

Over around 300 non-replaygained mp3 songs I tested, after applying a mp3gain of -3 dB, I found around 10 that still had some clipping, all of them old-dowloaded 128 Kbps mp3 or 64 Kbps compressed samples for this test. I found that this clipping was very little in all cases, with a worst case of a song with a few isolated clipped samples distributed all over the song, around 15 samples in this worst case. In the other worse case, I found a max. of 3 consecutive clipped samples, a couple of times in the song. I believe this clipping is little and distributed enough as to have no audible consequences at all.

Anyway, if you want to be extra safe, use a higher reduction (as said, be it by replaygaining the songs or by moving the DSP preamp to a higher value).

I'm being so stubborn with this because I strongly believe that the hard limiter is evil . Unless all original signal falls below the range of action of the limiter, that is, -6 dB below full scale, it will always distort the sound and make things audibly different. I produces a fuller and slightly coloured sound, but this is not an accurate way of playing music, unless desired. I think this is one of the main reasons why people say that Foobar2000 sounds different to Winamp. On the other hand, a little clipping as the one I've described, I think is not audible at all.

Sorry if I annoyed anyone with my previous posts.

I guess a quick summary of my views on the matter is what I posted at this thread (sorry if you already read it): http://www.hydrogenaudio.org/forums/index....=5770&st=25&hl=

AFAIK Peter is working on enhancing it with lookahead, so it only becomes active when it knows the signal is going to clip 'soon'. That would be the best of both worlds I think. No clipping -> no limiter.

"transparent look-ahead limiter"

This is pretty much what it will behave like, this is from the matlab project.
The limiter is transparent all the way until a clip or near-clip sample and it sounds pretty good when it clips too.

note! The "6dB Hard limiter" should really be called a "Soft clipping limiter" or "soft distortion limiter" since that's what it really is. it still clips, but does this in a soft manner.

now known as "advanced limiter"