Skip to main content ↑ ↓

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: The meaning of "Transition Band" (Read 6813 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

The meaning of "Transition Band"

Can you please tell me where can I find some information about transition band?
Why is it useful?
Why is it 15% of lowpass? (I mean LAME default setting here)
What should I expect if I change the default value? (LAME, again)
etc...
Caede te ipsum per murum.

The meaning of "Transition Band"

Reply #1 –
The transition band of a filter is a range of frequencies in which the filter 'transits' from passband to stopband. So called brickwall filters have very small transition band and are pretty hard to construct. There are problems like passband ripples near cutoff frequency and severe phase distortion. The wider the transition band, the easier is to get a good quality filter. But it eatlier starts to attenuate frequencies lower than cutoff. So, often one must make a good compromise.
If age or weaknes doe prohibyte bloudletting you must use boxing

The meaning of "Transition Band"

Reply #2 –
Quote from: eevan on
The transition band of a filter is a range of frequencies in which the filter 'transits' from passband to stopband. So called brickwall filters have very small transition band and are pretty hard to construct. There are problems like passband ripples near cutoff frequency and severe phase distortion. The wider the transition band, the easier is to get a good quality filter. But it eatlier starts to attenuate frequencies lower than cutoff. So, often one must make a good compromise.
I found your post very informative, thanks.

I also wanted to mention that most or all LAME versions after 3.94 "secretly" have a brickwall filter at 20 kHz (it uses the ATH formula and the output of the Modified Discrete Cosine Transform for this, chopping off all coefficients above 20 kHz), so this is active simultaneously with the polyphase highpass filter freq that LAME reports. And when I accidentally did some very primitive tests, encoding test tones near 20 kHz and checking the spectra visually (MDCT coefficients) I noticed pecularities in the encoded output as well (an imperfectly functioning brickwall filter).

I have no idea what impact, if any, these things have on audible sound quality, however.

The meaning of "Transition Band"

Reply #3 –
Quote from: Porcupine on
...I also wanted to mention that most or all LAME versions after 3.94 "secretly" have a brickwall filter at 20 kHz...


Wow! Those "evil" Lame developers "secretly" compressing files in a lossy way!!!!

Quote from: Porcupine on
... I accidentally did some very primitive tests, encoding test tones near 20 kHz and checking the spectra visually (MDCT coefficients) I noticed pecularities in the encoded output as well (an imperfectly functioning brickwall filter)...


Have you ever being able to HEAR any pecularities?

No. Otherwise you would be mentioning it here.

Quote from: Porcupine on
...I have no idea what impact, if any, these things have on audible sound quality, however.


If you couldn't hear any impact then it caused no impact.


This porcupine guy never learns!   

The meaning of "Transition Band"

Reply #4 –
> Wow! Those "evil" Lame developers "secretly" compressing files in a lossy way!!!!

I never said anything about Lame being evil, but secretly doing undocumented things is indeed very bad in my eyes.

> If you couldn't hear any impact then it caused no impact.

This statement is untrue. This kind of narrow thinking is one of the problems with most (not all) of the members on this board, in my opinion.

Just because I haven't heard any impact so far doesn't mean I wouldn't find a file where I heard an impact in the future. One cannot say for certain there is no impact on audible sound quality until all the songs in the world have been tested, each by all the people in the world. Listening tests are good, but they are not perfect. Whether or not someone is willing to base his decisions entirely on the listening tests conducted by one community of people, is up to each individual to decide. I, for one, am not.

The meaning of "Transition Band"

Reply #5 –
Quote from: Light-Fire on
Quote from: Porcupine on

... I accidentally did some very primitive tests, encoding test tones near 20 kHz and checking the spectra visually (MDCT coefficients) I noticed pecularities in the encoded output as well (an imperfectly functioning brickwall filter)...

Have you ever being able to HEAR any pecularities?

He won't. But definitely violating TOS. I can't beleieve he's surviving through a set of touchy discussions.


Quote
Just because I haven't heard any impact so far doesn't mean I wouldn't find a file where I heard an impact in the future.

How long it takes to be a person like that? As long as you are not listening, the day won't come. All tunings are for perceptual quality not for the incorrect objective measurement.

The meaning of "Transition Band"

Reply #6 –
Quote from: Porcupine on
...This kind of narrow thinking is one of the problems with most (not all) of the members on this board, in my opinion...


Porcupine. There is this forum that is almost like Hidrogenaudio antithesis. It is a place where all illogical people hang out together. You should go there. You will be happy there!!!

The meaning of "Transition Band"

Reply #7 –
Quote from: Porcupine on
Just because I haven't heard any impact so far doesn't mean I wouldn't find a file where I heard an impact in the future.

True. But with listening experience (your own and that of other people - there's a lot out there) there's progress at least though listening experience will always be restricted.

But with your theoretical reasoning you don't get forward a bit cause you have no idea what your considerations really mean speaking of listening relevance. And you don't realize that as you are moving in a world of lossy encoding, mp3 moreover with its special restrictions, you do have to accept an audio behavior which is necessarily imperfect the way you look at it (but nonetheless will be great for enjoying music).

That's why I really don't understand why you don't immediately move to the world of lossless encoding or very high quality wavPack or Optimfrog Dualstream lossy. This is the world where everything's fine for you. It'll take you less time to move there than the lot of time you spend with your kind of imperfections of mp3.
With your spectogram methodology you'll have no problem at all to find a setting for say wavPack lossy which will totally satisfy you.
lame3995o -Q1.7 --lowpass 17

The meaning of "Transition Band"

Reply #8 –
I do take listening experience (both of myself and others, but I weight heavily against others) into account also. But compared to most people I take theoretical reasoning and investigations into heavy account, yes.

> But with your theoretical reasoning you don't get forward a bit cause you have no idea what your considerations really mean speaking of listening relevance.

Hrm. Well, so far my personal encoding preferences have been decided according to the method: 1) Think a very long time about theoretically what is best, analyzing graphs, etc....2) encode with unusual parameters to satisfy my current theoretical ideas as to what may be best....3) listen as carefully as I can to everything I encode to make sure it's not any worse with my settings than typical settings (by trying to compare with the original WAV or mp3s made by other people or myself with more normal settings).

I dunno, I acknowledge my method is super dangerous but I think step 3) makes what I'm doing a lot safer. I think what bothers a lot of other people is that I put almost no weight into the listening experiences and habits of others. Well, that's true. And I openly admit it. Then again, how many people have been encoding with --noath? Probably not many, so there's not much evidence against me, I have to rely on myself to see if I did something bad or not.

Probably I should go back and encode with LAME 3.92 (or whatever version of old LAME). That way I wouldn't feel a need to specify any unusual commandline parameters whatsoever (Stereo is default on old LAME, and ATH "issues" dont seem to exist either) and no one would criticize me.

I acknowledge that lossy encoding is extremely imperfect, but I don't see why that should stop me from at least trying to get it as good as possible to satisfy my preferences.

> That's why I really don't understand why you don't immediately move to the world of lossless encoding or very high quality wavPack or Optimfrog Dualstream lossy. With your spectogram methodology you'll have no problem at all to find a setting for say wavPack lossy which will totally satisfy you.

I'm not sure the spectrogram methodology would help me with analyzing Wavpack lossy. It only really helps me with mp3 and nothing else, mainly because I think I am looking at a 576-point MDCT which exactly corresponds to the storage format of mp3, and makes things very visible. With Ogg and MP2 and AC3 and such, the spectrums on this don't look much different to me than the original WAV.

In any case, I noticed that Wavpack 4.41.0 final was just released, too. I'll try to hurry up getting to it, but it'll take me a while like I said earlier. I plan to have to play with it a LOT (and listen a LOT) before I start talking about it. Since I have absolutely no listening experience with that type of lossy compression, I have a lot of listening practice to do. I plan to extensively test how far I can push it (how low a bitrate), and settings (which you kindly gave me a lot of advice on, and which I've all saved for future reference), etc. Although in the end I will use a high bitrate of course, I should probably test with low bitrates to gain an idea how it works and sounds.

Also I have a slight reservation about Non-transform based forms of audio compression, which make me think I have to test Wavpack lossy all the more carefully. I still remember the old stuff like u-law, ADPCM, etc and those were awful (of course, those are zillions of times worse than Wavpack but still). And XP3 format (Playstation One audio compression) and ADI CRX (Dreamcast, GameCube, PS2 audio compression) might be in that category as well...and those sound awful to me too.

The meaning of "Transition Band"

Reply #9 –
Quote from: Porcupine on
I still remember the old stuff like u-law, ADPCM, etc and those were awful (of course, those are zillions of times worse than Wavpack but still).

Those were never meant for high quality coding of music material. When something is engineered, you have to stick to the goals and requirements. µ-law is just another quantisation function for PCM (like linear LPCM) used in telephony. ADPCM do compress, but is of no use for HI-FI.
If age or weaknes doe prohibyte bloudletting you must use boxing

The meaning of "Transition Band"

Reply #10 –
Quote from: Porcupine on
... encode with unusual parameters to satisfy my current theoretical ideas as to what may be best....3) listen as carefully as I can to everything I encode to make sure it's not any worse with my settings than typical settings ... I think what bothers a lot of other people is that I put almost no weight into the listening experiences and habits of others.

That's absolutely okay, you have to judge it on your own for the most important part. But without abxing your listening experience is worthless. Placebo is so strong.
It's also not wrong just to do careful listening if you just stick to usual settings as long as you don't hear a problem. Sounds like the best of common sense to me.
But the more you are going to do something like -k, -noath, the more you are obliged to do abxing with a lot of music to see that these settings don't make things worse if you really want to be into these things. Of course there's also nothing wrong finding a setting you are absolutely pleased with no matter what it's like. However if you're writing about it here you should start reporting about your abx tests.

Moreover there's a bit of more serious trouble out there in the mp3 and Lame world which has a total different character than what you're after. Look at AlexB's new problem sample or the samples of the sandpaper noise kind or tremolo kind, or the classical pre-echo and other problems. That's where the problems are. Most people decide not to care too much about them, and maybe that's wise to do. But if you're out for problems you should take care of these and not so much whether or not there's an undocumented lowpass at 19 kHz or so of which you don't even know whether that bothers you acoustically.

Quote from: Porcupine on
I'm not sure the spectrogram methodology would help me with analyzing Wavpack lossy.

Guess it won't as there won't be a problem.
There will be problematic samples as well for wavPack lossy when using low bitrate, so just take care that bitrate is not too low. With mp3 you're restricted to 320 kbps. With wavPack lossy you can go as high as you like and use very high quality mode on a PC. For a rather low bitrate you may prefer OptimFrog Dualstream. According to my very restricted experience it looks like OptimFrog when not using -ans gives the more consistent quality at lower bitrate, especially when it's up to encoding artificial (electronic) music.
lame3995o -Q1.7 --lowpass 17

The meaning of "Transition Band"

Reply #11 –
> That's absolutely okay, you have to judge it on your own for the most important part. But without abxing your listening experience is worthless. Placebo is so strong.

Hrm. Well, this is one of the big problems with my risky approach to audio encoding, as well as one limit on the usefulness of ABX testing in general. I am encoding the way I do because so far I *haven't* been able to ABX any differences. Placebo effect is not applicable in my case because I haven't been able to discern anything. Only a "reverse-placebo" effect may be occuring, but there's no good way to get around the "reverse-placebo" effect. The greatest worry I have is that I may have failed to discern something I might learn to discern in the future.

I've been encoding with -k --noath because in theory the worst I could do is waste 10% to 20% of my bitrate doing so. In theory this could be made up for by increasing my bitrate past 320 kbps if I ever heard a problem (although this is a problem since 320 kbps is the maximum mainstream-compatible bitrate for mp3). On the other hand, if I leave a freq cutoff at 19 kHz, the worst I could be doing is permanently throwing away a valuable part of the original information...not something that could be made up for in any other way. (I'm not saying those freqs are valuable for sure, just the worst-case possibility, right now I don't know either way). I would rather do the first case (as long as I can't ABX any differences) because it could be easily fixed (by increasing bitrate) at least in theory.

But in any case, using Lame 3.92 with default parameters (Stereo, "mystery" ATH on) is probably what I should use from now on. I haven't actually encoded anything new in a long time (other than my 5 messed up CDs I mentioned earlier). When I encode new stuffs I plan to use Lame 3.92 or Wavpack (which I already downloaded but will take me months to test before I make a decision whether or not I want to use it).

> It's also not wrong just to do careful listening if you just stick to usual settings as long as you don't hear a problem. Sounds like the best of common sense to me.

This is fine too, it's what most people do. Definitely much safer than what I do.

> However if you're writing about it here you should start reporting about your abx tests.

Yeah that's one of the things I don't like about here. If I write about things that don't directly have to do with ABX testing, or that cannot be supported easily with ABX testing, people get mad. Oh well. (this isn't against the TOS #8 the way it is written, but it seems to be an "unwritten rule" of this forum).

> Look at AlexB's new problem sample or the samples of the sandpaper noise kind or tremolo kind, or the classical pre-echo and other problems. But if you're out for problems you should take care of these and not...

I hear you. The thing is though regarding those problem samples, I wonder if they go away with a higher bitrate? If they do, that's great. If they don't though (which seems to be the case at least from what I've heard), there's nothing you can do about them either way so it does no good for me to worry about them. They'd be there whether I used normal settings or not. That's why I haven't been aggressively studying them.

In the case of AlexB's problem sample, one issue is that he can clearly ABX both the V2 JS and the V2 Stereo version from each other and the original, right? But then in that case it doesn't prove which one sounds better. It's subjective because he can distinguish both. He said that his higher bitrate tests may be forthcoming, though. I also said in that thread that I currently think LAME overcodes the M channel on Joint-Stereo VBRs by 20% extra allocated bits (which only supports what everyone has tested and heard). That only makes doing fair comparisons all the more difficult (perhaps impossible). Anyways, I really distrust ABX double-blind testing when used inappropriately...which is kind of what I feel this forum does (going 'overboard' with it is one way to utilize it inappropriately, as the forum doesn't recognize the limitations of what ABX testing can practically achieve). But if we want to discuss ABX testing's merits in more detail maybe we should do it in PM as it seems pretty off-topic now.

The meaning of "Transition Band"

Reply #12 –
Quote from: Porcupine on
...In the case of AlexB's problem sample, one issue is that he can clearly ABX both the V2 JS and the V2 Stereo version from each other and the original, right? But then in that case it doesn't prove which one sounds better.

AlexB has clearly stated that the joint stereo version sounds better.
I personally was interested only in the problem by itself (and Lame 3.98's behavior as opposed to 3.97's resp. ABR vs. VBR behavior). It's a real good sample which shows that choice of the encoder (and/or version) can be essential as well as relevant settings.
lame3995o -Q1.7 --lowpass 17