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: lame 3.94 regression with low bitrates? (Read 5221 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

lame 3.94 regression with low bitrates?

Because of this thread in these boards , I tested Lame at lower bitrates, concretely at 56kbps.

I started to play with a custom command, because I used lame 3.91 (jonh33 compile, which is bit exact to 3.90.2 but a bit faster) and it doesn't accept --alt-preset cbr values lower than 80.

Now, Since I saw some recommendations of --alt-preset cbr at 64kbps, I checked lame 3.94 to do the same tests, and I found that:

a) --alt-preset cbr 56 with 3.94 was supported, but was giving less quality than my preset with 3.91.
B) my custom command sounds horrible with 3.94 ( --mj -q2 -b56 --athtype 4 --nspsytune --lowpass 10000 --scale 0.95 )

[Edit: added "--scale 0.95" , recoded the files and updated them in the server ]

Here are the samples:

Lame 3.91 custom command
Lame 3.94a13 custom command
Lame 3.94a13 -aps 56
Sample in .flac (was resampled with ssrc to avoid lame resampling)

rename the .zip's to .mp3 (my host doesn't like .mp3's, no matter if they are legal or not)

Hope that this regression has a reason..

lame 3.94 regression with low bitrates?

Reply #1 –
..
edit: tryed to remove my post, but it seems that this is not possible

lame 3.94 regression with low bitrates?

Reply #2 –
Quote
..

how should I understand those ".." ?

I don't care that my commandline sound worse. Sure it is because the defaults have changed and maybe even the way some of them work.

The *real* problem is that  alt-preset sounds a bit worse than that not-really-wise commandline.

lame 3.94 regression with low bitrates?

Reply #3 –
Is that an isolated case, or is this constant for all files?

lame 3.94 regression with low bitrates?

Reply #4 –
Well, I am hearing an obvious difference: the levels

--preset 56 is using --scale 0.95 to prevent clipping, while your custom command line does not change the scale.

edit: it is not because of the scaling, and you updated your files with the proper scaling

lame 3.94 regression with low bitrates?

Reply #5 –
Quote
Well, I am hearing an obvious difference: the levels

I can confirm the problem noticed by [JAZ].
I've done a quick test (3.92 vs 3.94a13 ) with waiting.wav and applaud.wav at 56 Kbps (--alt-preset 56).
With waiting394a13.mp3 there is a sort of "ringing" that should indicate psycoacustics problems.
Try with applaud.wav, worsening is very easy to hear.

edit: deleted the last part of the post due to an error i've made.
WavPack 4.3 -mfx5
LAME 3.97 -V5 --vbr-new --athaa-sensitivity 1

lame 3.94 regression with low bitrates?

Reply #6 –
Could you compare it against 3.93.1 --preset 56 ?

lame 3.94 regression with low bitrates?

Reply #7 –
Quote
Could you compare it against 3.93.1 --preset 56 ?

The problem is present in all the old 394 alphas i own.
Even takehiro-lame.exe (3.93 alpha 2, Oct  9 2002) has the same problem.
3.93.1 stable is ok.
WavPack 4.3 -mfx5
LAME 3.97 -V5 --vbr-new --athaa-sensitivity 1

lame 3.94 regression with low bitrates?

Reply #8 –
[Edit : reedited the post for clarification]


"3.91 custom" and "3.93.1 --preset 56" sound quite similar (i preffer the custom one, because of the higher lowpass, but both have artifacts).

3.94 --alt-preset cbr 56 and 3.94 --preset 56 sounds similar too, and quite worse than the other two.


(I supplied the ".flac" sample in the first post, so you could do any tests you wanted)

lame 3.94 regression with low bitrates?

Reply #9 –
You got 1 point here (btw it is easier to hear with applaud.wav)
3.93.1 preset 56 is better than 3.94a14 preset 56 on those samples.

(I used the flac version to test)

lame 3.94 regression with low bitrates?

Reply #10 –
I would save me some time if someone could dig up 3.93 and test it on this sample.
I have a "feeling" that perhaps 3.93 could be ko in the same way as 3.94 on this sample.

lame 3.94 regression with low bitrates?

Reply #11 –
Found version 3.93 here (it starts to be difficult to find it! I think that's good  ) : http://click.alltheweb.com/go2/2/atw/1c527...r/lame-3.93.zip

--preset 56 sounded very similar to 3.93.1, (with both files), but not the same. In other words, the problem that 3.94 has, is not related to that (or at least, doesn't produce the same problems)

lame 3.94 regression with low bitrates?

Reply #12 –
Hi Again.

I think you are half right with the assumtion that 3.94 could have a problem similar to 3.93, but not the only one.

Here there are a few images with spectogram display, which shows two different problems.

The most notable one is big holes in the "high" frequencies, only present in 3.94 ( but hey.. here "high" means 8Khz...11Khz! it is not something one cannot hear ). This is the bug I pointed out.

The other one is some sort of blocking, which is not present in 3.91 but it is in 3.93, 3.93.1 and 3.94

Here there are the images (Click on them to view a bigger one) (The small ones are just 2KB each)
The frequency range is from 0 to 11Khz. The time range is aproximatedly between second 0.4 and second 0.8 (yep... 400milliseconds)

Original.


lame3.91 with custom command. Maintaining the original quite well (except for the filter).


lame3.94 --alt-preset cbr 56. Shows many holes in the "high" area, which are, of course, very noticeable.


lame3.93.0 --preset 56. In the small image is not much appreciable, but in the big one, you can see the "blocking" I was saying. It looks like skipping (block repeating).


lame3.93.1 --preset 56. Comparing this to the previous one, the blocking is still present, but there'a a small hole in the highs in 3.93 which is not present in 3.93.1. (at the beginning)


lame3.94 --preset56. Surprisingly, here you can see both, the highs holes, and the blocking!!!


When seeing this, I wanted to test 3.91 with --abr (because I hadn't done it yet), and got this:

lame3.91 --abr 56. The filter is lower (this is normal), and looks much like the cbr one, EXCEPT in the very beginning. Yes, the image is correctly set. There's a missing "beat". Anyway, this is just anecdotic.


Hope this is not as serious as it sounds..

Edit: If you compare the 3.94 --preset 56 (or --ap cbr 56) and the 3.91 --abr 56, you'll see that 3.91 filters where 3.94 starts to fail.
Might this indicate that somewhere in the code, this is hardcoded? (like if the codec thought "don't care about anything over 8Khz, because it will get filtered")

Edit 2: Investigated a bit more. The blocking ONLY happens with --(alt-)preset xxx . It does not appear using CBR, VBR nor even with --abr xxx .

Edit 3: Checking the 3.94 custom, I've seen that it has blocking too (just one) Click here to see the image. The blocking is just in the middle. The spectogram is completely messed up, just like it sounds, but this is another story.

lame 3.94 regression with low bitrates?

Reply #13 –
I checked the problem that appeared in 3.93 and was solved in 3.93.1, and we are not dealing with the same one (this one was related to short blocks, and this is not the case here)

On your spectrograms, there is something that I'm noticing with 3.94: the lower freqs. They seems to be more accurate than in 3.93.1, perhaps at the expense of less accurate highs.

Btw, I'd like to tell you that I appreciate your efforts. This really helps.

Edit: I am not able to test right now. If you wish, you could try "--preset cbr 56 --ns-bass -3" with 3.94.

lame 3.94 regression with low bitrates?

Reply #14 –
Note :
* Using alpha 14 now. (No difference so far)

--preset cbr 56 --ns-bass -3 gave a result very similar to --(alt-)preset cbr 56 (picture number 3 above), except it had one block in the middle again ( It looks like the next block is decoded ).
About the highs, they seem to be more accurate, without --ns-bass -3.

MP3 file

Image

lame 3.94 regression with low bitrates?

Reply #15 –
I found the problem. There was a bug in the code setting internal parameters which sometimes caused wrong noise shaping method to be used.
I'll correct it this weekend.

lame 3.94 regression with low bitrates?

Reply #16 –
Quote
I found the problem. There was a bug in the code setting internal parameters which sometimes caused wrong noise shaping method to be used.
I'll correct it this weekend.

Will it fix that problem?

http://www.hydrogenaudio.org/forums/index....ST&f=16&t=9044&
Vital papers will demonstrate their vitality by spontaneously moving from where you left them to where you can't find them.

lame 3.94 regression with low bitrates?

Reply #17 –
Both problems should now be fixed