I use "-Y" switch between V2 ~ V0
Is the quality worse below 16khz (between 0~16khz)

Yes, because without the -Y switch, the encoder is forced to use more bit below 16kHz than it believes are necessary, in order to encode above 16kHz at the quality which is believes is necessary.

So adding -Y decreases the quality below 16kHz, but only down to the level which the encoder believe is correct.

Or, to say it the usual way, without -Y information below 16kHz can be over coded (i.e. with more accuracy than is needed).

Of course, with -Y, information above 16kHz is under coded (less accurately than the encoder believes is necessary).

Cheers,
David.

thank you for your answer

then if I use "--lowpass" (example 17100Hz)
Is the quality worse below 17100Hz

This isn't what I've come to learn about the -Y switch. Lame help says it allows noise in the band above 16kHz to be ignored.

As an example, there is a great deal more high frequency content without the -Y switch when encoding a track with 20kHz of BW using Lame 3.97 -V1 --vbr-new. At the same time, the bitrate of the track without the -Y switch is significantly higher.

I find it difficult to make any conclusions about the quality below 16kHz.

It shouldn't be any worse. I recommend using --lowpass 16 instead of -Y.

edit: -Y should reduce quality below 16khz.. see my post below

If I remember correctly, the Y switch was highly recommended if you wanted smaller file size and wasn't too sensitive about higher frequencies. It doesn't cut as abruptly as the lowpass filter. I would therefore recommend using it over the lowpass.

Dibrom explains here:

http://www.hydrogenaudio.org/forums/index....;f=16&t=703



So far I can't get good abx results on vanilla -v2 vs -Y or --lowpass 16.

It should be BETTER below 16Khz because you aren't wasting bits on the stuff you probably can't hear (above 16Khz) anyway.

I can see why you'd say this if the bitrates were the same with or without -Y, but they aren't.

This seems to contradict your point also:

Yeah, see, I was right, as always.



not.

I just remembered that thread (have some of us really been here that long?!)

Cheers,
David.

The next question is, has the function of -Y changed since 2002?

reading the longhelp of lame 3.97 you will see no changes of the -Y switch.

Yeah, I joined a few months before that thread. I didn't know crap about lossy encoding then, and now I'm wondering if I do now, lol.

Me too!

Since it is VBR that is in question.. the quality below 16khz should remain the same since LAME's VBR = constant audible quality (or at least is intended to). So the only difference should be a reduced average bitrate, and bit-starved or dropped higher frequencies.

Now that I think about it, using -Y with -V 2 should actually reduce quality below 16khz. Let me (try) to explain why.

Lets say a certain frame is sized at 192kbps with -V 2.

This frame with -V 2 --lowpass 16 is 128kbps. (audio below 16khz should be audibly equal to the above)

This same frame is sized at 128kbps with -V 2 -Y (same size as above). The quality of this encoded frame will be worse below 16khz because -Y does NOT cut high frequencies. So now, instead of encoding all this bandwidth in the needed 192kbps frame, it is instead smashed into a 128kbps frame, thus it is reduced quality because the high frequencies take bits away from the lower frequencies. (this is why I am discusted by the fact that -Y is actually used at all, instead of just being an experimental switch)

FWIW, -Y only works with VBR, it does not work with ABR or CBR.

This does not mean that it will always encode them, either.

Have you actually seen this in practice or are you making this up?

It's clearly a hypothesis, and one he can only make by not reading the thread properly!

The -Y switch measurably (and, ?to some people?, audibly) does exactly what it was designed to. There is no way it should be removed.

Cheers,
David.

Well, several of us are there since a lot more time (don't you?). We come from r3mix.net, where we came from vqf.com, where we came from mp3.com, where we came from dmusic.com...ah, the good old days of winplay3, l3enc, nad, sonique, k-jofol...

Ah, you got me - I was never at dmusic.com.

But yes, we're getting old!

Cheers,
David.

Correct, but it since they aren't always dropped, the quality can be reduced due to -Y being used in VBR.

I have tested -Y quite a lot (which is why I hate it).

How much this effects audible quality at -V 3 and -V 4 (the only VBR modes using -Y with freqs higher than 16khz).. I don't know.. but IMHO those modes use a high enough bitrate for it not to matter much (at least not to me, lower bitrates like 112kbps makes the problem more audible). Custom compile lame without -Y on -V3 and you will find the average bitrate very close to -V2 (because the frequencies above 16khz are taken into account when encoding).

In lame's help it says (something like):
-Y allows lame to ignore noize in sfb21 like in cbr
This means (please correct me if I'm wrong) that lame ignores any sound above 16khz when calculating the frame size in VBR. The noise is still there, but lame pretends it's not. Thus lame only calculates the frame size for 16khz of audio bandwith. Which leads to my conclusion that since -Y does not necessarily cut freqs above 16khz, lame encodes more than 16khz of bandwidth into a frame intended for only 16khz of bandwidth. Which results in lower quality (at least conceptually).

Every test I have done with -Y has proven this to be true.

I still remember testing the -Y switch at r3mix with the macabre sample and its triangles at r3mix.net
Today i have problems making clear statements about this

We ARE old (at least i feel so) nonetheless i am still surprised i can still hear a 18.5kHz sweep and a strong 16kHz lowpass at the age of 38!

Cheers!

I don't doubt your ability to ABX differences, but this is a far cry from explaining your theories. Nothing you've said is consistent with what has been written by the reputable experts in this forum.

There is no difference in general. I do have a problem sample for LAME and in that case plain v2 is a little better as more bits are going under 16k - not because they are needed. I can't normally abx -Y vs plain v2 and stuff above 16k is encoded as long as its not gonna cause bitrate bloat.

Slightly off topic, but the V3 setting seems to not use sfb21, so, is it using the -Y switch by default, or some other method to reduce bitrate bloat above 16khz.

With Lame 3.97, at least, "-V3 --vbr-new" and "-V3 --vbr-new -Y" deliver exactly the same results.

V3 and lower quality use the -Y switch
http://wiki.hydrogenaudio.org/index.php?ti...cal_information

Thanks for the link. I feel stupid for not looking there first

noise not equal sound.
Concretely, it says it doesn't care about encoding *accurately* (noise -> innacuracies ) those frequencies.
I don't get how you related "ignore noise" with "pretend to put an elephant in a t-shirt pocket", if you get what i mean.

Because that is what all of my testing has shown.

Unless you can demonstrate a situation where Lame has packed more bandwidtch into a frame of the same size or smaller using -Y with the rest of the settings being the same, you will not be able to support your theory.

Based on Dibrom's statements and the simple fact that significantly less information above 16kHz gets encoded, I don't believe you'll ever be able to do this.

proof (screen shots are from analfreq 1.8 using left channel, test files are available there too)

The -Y situation can cause an artifact that I would describe as 'chirping', because sometimes there is sound/noize above 16khz. And when there is sound up there, it gets smashed into a frame* intended for 16khz, not more.


- 1kbps is not enough for me to consider the -Y and --lowpass 16 encodes to be very different.

Don't like my proof? Do your own tests and prove me wrong.

*if this is not technically the correct term, sorry and please correct me

VBR doesn't work this way. For example with VBR NEW it goes like this:
- psymodel says how much noise is allowed per scalefactor band
- quantization step sizes are calculated which would result in lower or equal noise in each band
- global gain and scalefactors are selected to reach Layer3 specific constrains (like no sfb21 scalefactor, limited range of quantization step sizes, etc)
- frame gets quantized and encoded
- if it fits into available bits (a 320 kbps frame if you don't toy around with -Bxx switch), then that's it
- else quantization step sizes are adjusted until encoded data fits into available bits.


Using the -Y switch, anything in the sfb21 range (over 16 kHz) gets encoded "as-is" using the global quantization step size.

gameplaya I don't understand this graph... can you explain me please what the blue and the red marked lines are for? and the numbers for the axis, what are they for?

...maybe also ask him if he knows what a frame is.

Sorry, but I do not understand the purpose of this test...
What are you trying to prove??
I hope you are not taking conclusions like "this encoding method is better than this other because the graph has more high frequencies".

If -Y is used, is noise above 16khz take into account here? (or am I missing something here?)
I thank you for the explanation, but I don't think it has any affect on how the effects of -Y are observed.

red=average, blue=real-time, y-axis is amplitude, x-axis is frequency. If you want a better look at what is going on (in my test), you should get a copy of analfreq 1.8 and play around with it. The picture does a poor job of showing how much stuff there is above 16khz for my test file.


I am trying to prove that -Y is bad. I am not drawing any conclusions from how the graphs look, they are merely there to show what is going on.

I added some abx test results to that little page of mine. There are audible differences between the 3 encoded samples.

I have yet to determine how much of an audible effect -Y has on the sound below 16khz (as is the whole point of this... working on it... I think lower bitrates would be easier, but -V9 is practically worthless to test this in recent LAME)

Doesn't what I am saying make sense? (at least conceptually)
- 2 files are pretty much the same average bitrate
- one of them has more bandwidth than the other
- higher frequencies are harder to encode and take more bits
- to have more bandwith in the same size.. the lower freqs must lose some bits to the high ones
= therefore I have to conclude that the file with more bandwidth, must be lower in quality (below 16khz) than the other file

Am I wrong? (explain, with encoding tests that can be done)


Now I am off to play Gunz: The Duel