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: 3.94 alpha, Presets, Takehiro's Stable Branch, Etc... (Read 25884 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

3.94 alpha, Presets, Takehiro's Stable Branch, Etc...

We can continue discussion in this thread..

Keep it useful and objective people.  Negative comments (telling someone to shut up) or comments which are off topic (such as how to handle interaction with the lame-dev) will not be tolerated.

The only thing that belongs in this thread is discussion about problems with this build, test results, ways to improve it, or suggestions for release.

3.94 alpha, Presets, Takehiro's Stable Branch, Etc...

Reply #1 –
Dibrom,

that's all I found today:

------------------------------>8------------------------------

taking.wav


lame_p1 --preset fast standard

ABX: 16/16
There's a consistant ringing in the left channel; at the 2 sec mark where the first tambourine-like sound appears, it is most obvious.


lame_p1 --preset standard

ABX: 15/16
Same as with the fast preset, but the ringing is not as harsh.

------------------------------>8------------------------------

That's the only sample in my small personal collection I've found the new compile to have trouble with so far. It's available here. I'll keep searching and testing though.

BTW, if the encoder speed stays like this, that's one less argument against using LAME VBR. We're winning!

CU

Dominic

Edit: The download works now. Funny Opera wants to save it as .html though...

At least I'm not the only one having trouble with free web hosts *cough*

3.94 alpha, Presets, Takehiro's Stable Branch, Etc...

Reply #2 –
Thanks Volcano.

Can you test this sample with 3.92 and see how it compares?  Thanks..

3.94 alpha, Presets, Takehiro's Stable Branch, Etc...

Reply #3 –
I've uploaded most of the test samples that were used for the 3.90.2 --alt-preset testing to this directory:

http://static.hydrogenaudio.org/extra/samp...3-Test_Samples/

I have a few more samples not on me at the moment which I'll be uploading later, and then I'll probably zip them all into 1 big file for people to download also..

Some of these samples were used for tuning, some of them are samples that 3.90.2 had problems with (like the dog whistle clip, sopha, and save the birds, short, etc... most of which I haven't tested with Takehiro's 3.93 branch yet).  Some are also just some random difficult samples I had laying around..

3.94 alpha, Presets, Takehiro's Stable Branch, Etc...

Reply #4 –
I've uploaded ICL compiles of Takehiro's branch with the new presets (along with the sources) at the following urls:

Edit:  Check below a few posts below for another updated binary as well...

http://static.hydrogenaudio.org/extra/LAME...15_2002-bin.zip (binary)

http://static.hydrogenaudio.org/extra/LAME...15_2002-src.zip (source)

You can use these for testing.

The presets are more or less the same as in the previous compile I released.. a few things are different.  Also, there's probably lot of room left for tuning still, so test away

3.94 alpha, Presets, Takehiro's Stable Branch, Etc...

Reply #5 –
Hrmm.. one of the changes Takehiro made to his new branch affected the short block switching thresholds.  I believe he made this change to lower bitrate some, but at least in some impulse (ravebase, etc) clips it has made things a little worse (though in general performance on impulse clips is improved, short especially shows *big* improvements).  I've changed it back in a new build I'll be uploading shortly..

3.94 alpha, Presets, Takehiro's Stable Branch, Etc...

Reply #6 –
Updated lame with more aggressive short block thresholds can be found here:

(Removed due to some tuning issues..)

3.94 alpha, Presets, Takehiro's Stable Branch, Etc...

Reply #7 –
@takehiro:

Can you take a look at the drone_clip.wav sample in the directory I provided?

Even with the more aggressive short block thresholds, it still has some problems.  If you lower the gfc->nsPsy.attackthre to 1.1 then the sample sounds much better, but obviously that also increases the bitrates on other samples tremendously.  Maybe you can take a look?

Also, the Sophia2.wav and Save The Birds.wav clips still have the same problem as in 3.92... there is some distorted noise on the vocals.  Like the vocals sound "raspy".

And the dogwhistle.wav clip has the same problem as in 3.92 also... the tone is distorted sounding.  The problem didn't exist with vbr-new under 3.92.  I haven't tried vbr-new under 3.93 though.

Btw, decreasing the gfc->nsPsy.attackthre from 8.0 to 3.5 (like it was in 3.92) fixes a problem on the ravebase sample at ~31 seconds were there is a bit of a stutter.. and then another "blip" sound around ~34 seconds in the left channel.  With the gfc->nsPsy.attackthre lowered to 3.5 this sample sounds almost perfect now... the other parts that were distorted before seem to be fixed by the new impulse code.

Also, I believe using 3.5 instead of 8.0 might also reduce pre-echo on castanets some... JohnV said earlier that your branch showed a little bit increased pre-echo on the first 4 attacks of that sample compared to 3.92.

I think that dogwhistle.wav, Sophia2.wav, Save The Birds.wav, and drone_clip.wav are probably the biggest problem samples I can think of at the moment that still remain.  Short.wav was a big problem with 3.92 but that sample is fixed in 3.93 thanks to your new impulse code

3.94 alpha, Presets, Takehiro's Stable Branch, Etc...

Reply #8 –
Hrmm... looks like a gfc->nsPsy.attackthre of 3.5 may not be optimal in the new code in all cases.  Although it fixes a problem in Ravebase, it also causes a very bad skip in death2.wav ~5 seconds...

Looks like some more tuning is still needed in that regard.

3.94 alpha, Presets, Takehiro's Stable Branch, Etc...

Reply #10 –
Dibrom,

here are some more results (not many):

http://free.pages.at/volcano/temp/results.txt

The results from above are included as well, for a better overview.
That sample doesn't actually seem to be bad. I'll try to find more...

Edit: Oh shit! I wasn't using the latest compiles, I had completely forgotten about them! Damn it. I hope the results still have some value

CU

Dominic

3.94 alpha, Presets, Takehiro's Stable Branch, Etc...

Reply #11 –
Quote
Edit: Oh shit! I wasn't using the latest compiles, I had completely forgotten about them! Damn it. I hope the results still have some value

Well, the "latest" compiles have some problems... I should probably take them down for now.

3.94 alpha, Presets, Takehiro's Stable Branch, Etc...

Reply #12 –
One more question: Do the presets scale down the volume, do we have to use --scale 1? I didn't until now, but the thread on --scale just reminded me...

3.94 alpha, Presets, Takehiro's Stable Branch, Etc...

Reply #13 –
Quote
One more question: Do the presets scale down the volume, do we have to use --scale 1?

To my knowledge, the VBR presets never used any scaling.

3.94 alpha, Presets, Takehiro's Stable Branch, Etc...

Reply #14 –
--alt-preset medium is set to use qval=2, which slows it down immensely. It obviously should be set to qval=3 like the rest.

3.94 alpha, Presets, Takehiro's Stable Branch, Etc...

Reply #15 –
Quote
--alt-preset medium is set to use qval=2, which slows it down immensely. It obviously should be set to qval=3 like the rest.

--alt-preset medium is gabriel's preset.... the one I created is dm-medium.

3.94 alpha, Presets, Takehiro's Stable Branch, Etc...

Reply #16 –
More results here.

I'm not making this "waste" sample available at the very moment. It may still be lying around the webserver of the person who found it (xoggy), which is what I'm trying to find out first before I send large files (56K...). You'll have to wait until mid-day (CET).

All I can tell you for the moment is that it's from a Metallica song called "Wasting my hate" IIRC. It contais a "high" e-bass tone on the left channel and percussion on the right channel that sounds like hi-hat (or perhaps *is* hi-hat, I'm not at all familiar with all the expressions ). It introduces a rumbling noise similar to what we know from ff123's 128kbps setting with ns-bass improperly set.

Dibrom: Can I send you the file via email if I run into trouble uploading it to a free server (filesize limits etc.)?

CU

Dominic

3.94 alpha, Presets, Takehiro's Stable Branch, Etc...

Reply #17 –
Preset medium is there because it seems to work correctly, and I though that it was better than nothing.

But if you come up with a better one, I'd be happy to see this one replaced by a better one.

Btw, I'd really like to know your thoughs about this preset.

3.94 alpha, Presets, Takehiro's Stable Branch, Etc...

Reply #18 –
Quote
Preset medium is there because it seems to work correctly, and I though that it was better than nothing.

But if you come up with a better one, I'd be happy to see this one replaced by a better one.

Btw, I'd really like to know your thoughs about this preset.

Is preset medium the -aps -Y?
In that case it will most likely have to be replaced because IIRC Dibrom's lower bitrate --alt-preset uses the 16kHz cutoff together with further tweaks of his own.

3.94 alpha, Presets, Takehiro's Stable Branch, Etc...

Reply #19 –
No, it's not "standard -Y"

3.94 alpha, Presets, Takehiro's Stable Branch, Etc...

Reply #20 –
waste.wav is available here:

http://www.ripncode.de/waste.zip

Please be graceful on the server, it ain't mine.  Also, the sample's not going to be on there for *much* longer.

CU

Dominic

3.94 alpha, Presets, Takehiro's Stable Branch, Etc...

Reply #21 –
It seems that dm-medium is a little too biased toward transparency. This is not a bad goal; however, since we are looking for recognizably smaller files versus standard, this profile might need more (slight) degradation.

Malory - Outerbeats - 01 - Lake Of Doubts

standard: 198.1kbps, LR: 37.52%, MS: 62.48%
dm-medium: 180.6kbps, LR: 37.52%, MS: 62.48%
medium: 154.2kbps, LR: 23.93%, MS: 76.07%
mppenc 1.13g, standard: 166.4kbps

Malory - Outerbeats - 02 - Xirius Polar Station

standard: 195.1kbps, LR: 57.06%, MS: 42.94%
dm-medium: 191.8kbps, LR: 57.06%, MS: 42.94%
medium: 161.5kbps, LR: 31.11%, MS: 68.89%
mppenc 1.13g, standard: 169.0kbps

I think that dm-medium should use a different stereo mode than standard.

If you consider, insane - 320kbps, extreme - 256kbps average, standard - 192kbps average, the lower-bitrate profiles should probably resemble: medium - 160kbps average, radio/portable - 128kbps average. From this perspective, Gabriel's medium profile works well bitrate-wise. However, casual listening says dm-medium sounds marginally better. Where we want to be is probably somewhere in the middle.

3.94 alpha, Presets, Takehiro's Stable Branch, Etc...

Reply #22 –
Of course dm-medium is barely tuned at all yet... I only just created it and right now I'm more concerned with getting the remaining files which cause big problems for aps fixed first.

However, --alt-preset portable is probably more comparable to Gabriel's medium preset.  That's the bitrate range that it's targeting anyway.  In one of my other posts I believe I mentioned that the bitrate range is supposed to roughly go as:

standard = 200kbps
dm-medium = 180kbps
portable = 160kbps
dm-radio = 140kbps

Below 140kbps I believe it will be very hard to reach decent quality with VBR without very clear and offensive artifacts.  It might be possible, but it won't be easy..

Again, the new presets will still need more tuning, but as you can see, dm-medium isn't really meant to be in the same bitrate range as gabriel's medium.

3.94 alpha, Presets, Takehiro's Stable Branch, Etc...

Reply #23 –
I've fixed the problem in the ravebase sample now by tweaking Takehiro's impulse code a little bit.. I'll be uploading a new build later tonight with this tweak.  I'll also try to tune the other presets a little more if I have some extra time..