IPB

Welcome Guest ( Log In | Register )

5 Pages V  « < 2 3 4 5 >  
Closed TopicStart new topic
lame3100k - bringing constraint VBR to Lame
[JAZ]
post Jul 28 2013, 10:58
Post #76





Group: Members
Posts: 1710
Joined: 24-June 02
From: Catalunya(Spain)
Member No.: 2383



Using both settings in the way you do, does what you want, but so far, the only reasoning to do so is visual, not aural. That's why i asked the question. Using it on the highest setting with the intention to retain more from the original signal might be ok, if the DAC and the resampler used (if needed) for playback do the job properly smile.gif

The problem that motivated the creation of the -Y switch was that treating the last band (16Khz..22Khz) the same way than the rest, caused an increase of bitrate, not because of content in that band, but because it forced all bands to increase precision. And that's fine (in terms of quality), except when there is not enough bitrate (since mp3 is limited to max 320).
Go to the top of the page
+Quote Post
halb27
post Jul 28 2013, 11:08
Post #77





Group: Members
Posts: 2414
Joined: 9-October 05
From: Dormagen, Germany
Member No.: 25015



QUOTE ([JAZ] @ Jul 28 2013, 11:58) *

... And that's fine (in terms of quality), except when there is not enough bitrate ....

And this does happen on occasion when using -V0 or similar. Using -Y or a lowpass decreases the number of these events.
Sure it's always a trade off between this and HF behavior, but as long as a person isn't affected by HF behavior with real world music IMO it's better restrict extreme HF behavior a little bit.
But it's nothing to really argue about as anybody can use -Y or --lowpass according to personal preference.


--------------------
lame3100m --bCVBR 300
Go to the top of the page
+Quote Post
greynol
post Jul 28 2013, 13:43
Post #78





Group: Super Moderator
Posts: 10000
Joined: 1-April 04
From: San Francisco
Member No.: 13167



Reasons behind personal preferences matter here.


--------------------
Your eyes cannot hear.
Go to the top of the page
+Quote Post
halb27
post Jul 28 2013, 16:04
Post #79





Group: Members
Posts: 2414
Joined: 9-October 05
From: Dormagen, Germany
Member No.: 25015



Hard facts behind personal preference is welcome most.
Do you know of such for deciding upon -Y or -lowpass?


--------------------
lame3100m --bCVBR 300
Go to the top of the page
+Quote Post
greynol
post Jul 28 2013, 16:17
Post #80





Group: Super Moderator
Posts: 10000
Joined: 1-April 04
From: San Francisco
Member No.: 13167



Have I stated a preference for one or another that was not based on DBT?


--------------------
Your eyes cannot hear.
Go to the top of the page
+Quote Post
NePaC
post Jul 28 2013, 16:41
Post #81





Group: Members
Posts: 16
Joined: 9-June 13
Member No.: 108568



@LedHed8
Here are my ABX test results using --bCVBR 192 and I could successfully abx the files.
CODE
16:50:07 : Test started.
16:50:39 : 01/01 50.0%
16:53:09 : 01/02 75.0%
16:53:43 : 02/03 50.0%
16:55:39 : 03/04 31.3%
16:57:24 : 04/05 18.8%
17:00:53 : 05/06 10.9%
17:05:17 : 06/07 6.3%
17:07:58 : 07/08 3.5%
17:11:43 : Test finished.

I have to say it gave me more trouble as my last test with the previous version. Normally if you aren't doing abx tests you probably won't notice the difference for most songs at all (myself included) I think.

@halb27
Little question regarding --bCVBR x. I want to understand how it is decided which n for -Vn is used when for example x = 260? And is the same n value used for the other music too?
Go to the top of the page
+Quote Post
halb27
post Jul 28 2013, 18:43
Post #82





Group: Members
Posts: 2414
Joined: 9-October 05
From: Dormagen, Germany
Member No.: 25015



--bCVBR 230 is -V1 --cvbr 5,
--bCVBR 249 is -V0.75 --cvbr 4.25,
--bCVBR 266 is -V0.5 --cvbr 3.5,
--bCVBR 278 is -V0 --cvbr 3.
Intermediate --bCVBR values according to linear interpolation.


--------------------
lame3100m --bCVBR 300
Go to the top of the page
+Quote Post
LedHed8
post Jul 29 2013, 05:07
Post #83





Group: Members
Posts: 38
Joined: 1-June 13
Member No.: 108428



Until you have objective evidence that this mod performs better than the others at producing transparency in a double-blind test, we are not interested in unsubstantiated suspicions about how it sounds, which violate TOS #8.
QUOTE


db1989, my apologies to you, all forum members, and forum guests for my TOS #8 violation. BTW, your photo did make me laugh aloud. laugh.gif I knew as soon as I saw it that I had crossed that line and I thought that was a pretty fun way to point it out. Beyond thanking halb27 for his work, the rest of my post ought not have been made. Forgive my over-exuberance.

NePaC, thank you for your abx test. I'm hopeful that my test has different results. I'm thinking that your ears are better than mine since I can't normally abx std lame V4, or iTunes/qaac 160cvbr. Vorbis (my portable preference for several reasons) tends to be a different animal for me, and I won't make further comment as to why in order to avoid another TOS #8 violation. I'm not particularly good at detecting pre-echo issues. However, I believe certain tonal/timbre situations are easier for me to identify.
Go to the top of the page
+Quote Post
halb27
post Jul 29 2013, 09:30
Post #84





Group: Members
Posts: 2414
Joined: 9-October 05
From: Dormagen, Germany
Member No.: 25015



I'd like to put things a bit into perspective.
Using bitrates ~200 kbps on average music is usually transparent, also with standard Lame.
It should be mentioned that the final quality is achieved by standard Lame in the first place which simply is great.
3100l can help on rare issues, and due to it's IMO negligible bitrate increase I think it's worth while using it. But it doesn't work miracles. At ~200 kbps there are audible issues left where lame3100l doesn't help.
Even with the highest quality settings overall transparency cannot be achieved. However the remaining issues are so small that at least to me they are negligible. But getting that close to overall transparency requires bitrates >250 kbps.

BTW if an ABX test takes 20 minutes for a 7/8 result it shows of course that the sample isn't transparent but it also shows that it is so close to it that I personally wouldn't care in real world listening situations especially when accompanied by corresponding remarks.


--------------------
lame3100m --bCVBR 300
Go to the top of the page
+Quote Post
NePaC
post Jul 29 2013, 12:41
Post #85





Group: Members
Posts: 16
Joined: 9-June 13
Member No.: 108568



QUOTE (LedHed8 @ Jul 29 2013, 06:07) *
NePaC, thank you for your abx test. I'm hopeful that my test has different results. I'm thinking that your ears are better than mine since I can't normally abx std lame V4, or iTunes/qaac 160cvbr. Vorbis (my portable preference for several reasons) tends to be a different animal for me, and I won't make further comment as to why in order to avoid another TOS #8 violation. I'm not particularly good at detecting pre-echo issues. However, I believe certain tonal/timbre situations are easier for me to identify.

You just have to add your ABX results under your claim. Didn't you plan to do one?
QUOTE (halb27 @ Jul 29 2013, 10:30) *
BTW if an ABX test takes 20 minutes for a 7/8 result it shows of course that the sample isn't transparent but it also shows that it is so close to it that I personally wouldn't care in real world listening situations especially when accompanied by corresponding remarks.

That's why I wrote two sentences, especially the last one, under the ABX test, to give my thoughts about the results. My conclusion was practically the same like what you wrote above, even so I could tell the difference, the time it took me to do that was to long for me, because the problem part was quite thin.
And thanks for telling me about bCVBR. Found it later in the code and noticed that I still had your first version of the 3100l version.
Go to the top of the page
+Quote Post
greynol
post Jul 29 2013, 13:29
Post #86





Group: Super Moderator
Posts: 10000
Joined: 1-April 04
From: San Francisco
Member No.: 13167



QUOTE (NePaC @ Jul 29 2013, 04:41) *
You just have to add your ABX results under your claim. Didn't you plan to do one?

No amount of ABX results will excuse that claim as it was made.


--------------------
Your eyes cannot hear.
Go to the top of the page
+Quote Post
LedHed8
post Jul 30 2013, 00:55
Post #87





Group: Members
Posts: 38
Joined: 1-June 13
Member No.: 108428



QUOTE
You just have to add your ABX results under your claim. Didn't you plan to do one?


Yes, I intend to as soon as I have time available.
Go to the top of the page
+Quote Post
halb27
post Jul 30 2013, 20:05
Post #88





Group: Members
Posts: 2414
Joined: 9-October 05
From: Dormagen, Germany
Member No.: 25015



I'll be on holidays for the next three weeks. I probably won't have access to the internet.
So if questions will come up for me during this period please be a little patient.


--------------------
lame3100m --bCVBR 300
Go to the top of the page
+Quote Post
NePaC
post Jul 31 2013, 00:58
Post #89





Group: Members
Posts: 16
Joined: 9-June 13
Member No.: 108568



QUOTE (greynol @ Jul 29 2013, 14:29) *
QUOTE (NePaC @ Jul 29 2013, 04:41) *
You just have to add your ABX results under your claim. Didn't you plan to do one?

No amount of ABX results will excuse that claim as it was made.

Wanted to say that he should do one before posting anything about what he claimed before, was mainly referring what db1989 wrote under the second quotation. You are probably referring to the first quotation, so I apologize if I didn't make it more clear.
Go to the top of the page
+Quote Post
greynol
post Jul 31 2013, 06:47
Post #90





Group: Super Moderator
Posts: 10000
Joined: 1-April 04
From: San Francisco
Member No.: 13167



One ABX test isn't going to elevate the status of the 99.9% figure above suspicion either.

Let's face it, that post belongs in the recycle bin.

This post has been edited by greynol: Jul 31 2013, 06:51


--------------------
Your eyes cannot hear.
Go to the top of the page
+Quote Post
eahm
post Jul 31 2013, 21:31
Post #91





Group: Members
Posts: 884
Joined: 11-February 12
Member No.: 97076



First time testing this extension sorry for the stupid question: are i, j and k different extensions/versions or they go alphabetically so k is the new one?

Thanks.

This post has been edited by eahm: Jul 31 2013, 21:31
Go to the top of the page
+Quote Post
Propheticus
post Jul 31 2013, 21:50
Post #92





Group: Members
Posts: 218
Joined: 10-September 11
Member No.: 93615



He posted the l (L) version a page back. I think it's alphabetical version lettering.
Go to the top of the page
+Quote Post
eahm
post Jul 31 2013, 21:51
Post #93





Group: Members
Posts: 884
Joined: 11-February 12
Member No.: 97076



Perfect, thanks.
Go to the top of the page
+Quote Post
SubV
post Aug 1 2013, 00:00
Post #94





Group: Members
Posts: 24
Joined: 27-November 08
From: Kursk, Russia
Member No.: 63333



I've compiled the Lame 3.100l with SSE2 optimizations enabled (which results in faster encoding); this version also features the default lowpass value increased to 19500 Hz.

You can download it here.

This post has been edited by greynol: Aug 1 2013, 23:13
Reason for edit: Link removed; the author does not appear to endorse your modification of the lowpass value.
Go to the top of the page
+Quote Post
Kamedo2
post Aug 1 2013, 03:27
Post #95





Group: Members
Posts: 168
Joined: 16-November 12
From: Kyoto, Japan
Member No.: 104567



QUOTE (SubV @ Aug 1 2013, 08:00) *
I've compiled the Lame 3.100l with SSE2 optimizations enabled (which results in faster encoding); this version also features the default lowpass value increased to 19500 Hz.

Was the SSE related bug fixed?
http://www.hydrogenaudio.org/forums/index....98383&st=25
Go to the top of the page
+Quote Post
SubV
post Aug 1 2013, 04:13
Post #96





Group: Members
Posts: 24
Joined: 27-November 08
From: Kursk, Russia
Member No.: 63333



QUOTE (Kamedo2 @ Aug 1 2013, 06:27) *

I don't think this is a bug - just different representation of floating point variables. The x87 compiler uses 64-bit registers, and SSE2 uses 80-bit. Different compilers produce different results, too.
Go to the top of the page
+Quote Post
LedHed8
post Aug 1 2013, 04:44
Post #97





Group: Members
Posts: 38
Joined: 1-June 13
Member No.: 108428



QUOTE (greynol @ Jul 31 2013, 00:47) *
One ABX test isn't going to elevate the status of the 99.9% figure above suspicion either.

Let's face it, that post belongs in the recycle bin.


Agreed... It was stupid. sad.gif
Go to the top of the page
+Quote Post
[JAZ]
post Aug 1 2013, 10:49
Post #98





Group: Members
Posts: 1710
Joined: 24-June 02
From: Catalunya(Spain)
Member No.: 2383



QUOTE (SubV @ Aug 1 2013, 05:13) *
The x87 compiler uses 64-bit registers, and SSE2 uses 80-bit.


I am sure you just make a mistake when writing it, but it's the other way around. (SSE2 less precise than x87).

Also, let's not forget that float is 32bit (and uses SSE operations), double is 64bit (and uses SSE2 operations), and that the 80-bit of x87 is only for the intermediate calculus of the value, because the values are still stored in either float or double.
Go to the top of the page
+Quote Post
SubV
post Aug 1 2013, 18:01
Post #99





Group: Members
Posts: 24
Joined: 27-November 08
From: Kursk, Russia
Member No.: 63333



QUOTE ([JAZ] @ Aug 1 2013, 13:49) *
I am sure you just make a mistake when writing it, but it's the other way around. (SSE2 less precise than x87).

Of course you're right. But if you compare different Lame builds, you'll find out that most of them give different output. Btw, the Intel C++ builds from Rarewares are SSE2 optimized, too.
Go to the top of the page
+Quote Post
IgorC
post Aug 1 2013, 22:10
Post #100





Group: Members
Posts: 1506
Joined: 3-January 05
From: Argentina, Bs As
Member No.: 18803



SubV, how are You?

It should be better for all if You coordinate all acitivity with halb27 as he is developer of this LAME extention. A development of lossy codecs can be very picky. Even such apparently harmless compilation with SSE2 optimization can cause a real mess. Don't mention to touch some defaults setting like lowpass.

Sorry, but I wouldn't recommend your build (not even for test) as it had no revision of main developer. It's his (halb27) codec.

This post has been edited by IgorC: Aug 1 2013, 22:12
Go to the top of the page
+Quote Post

5 Pages V  « < 2 3 4 5 >
Closed TopicStart new topic
1 User(s) are reading this topic (1 Guests and 0 Anonymous Users)
0 Members:

 



RSS Lo-Fi Version Time is now: 20th April 2014 - 03:31