LAME 3.97a4, How safe is it? Options

[LoFiYo]

I'd prefer adventurous people to test the next alpha version (ie the next 3.97a not yet available)

The current alpha 4 seems stable enough (it never crashed on any of the many samples I have encoded), and to me usually sounds the same as 3.96.1 and sometimes better. My question is, how safe is this alpha from a technical standpoint? Is it so experimental that no one should ever use it for anything serious? Or is it technically more stable than 3.96.1 (like 3.96a was technically more stable than 3.95.1 at one point)?

Also, I am curious about the upcoming alpha version

[Gabriel]

If you run the alpha executable, there is a warning displayed that states that alpha versions should only be used for experimentations.
This warning is real, it is not just an eye-candy.

When I say "test" I really mean testing so I can have some feedback.

[westgroveg]

If you run the alpha executable, there is a warning displayed that states that alpha versions should only be used for experimentations.
This warning is real, it is not just an eye-candy.

When I say "test" I really mean testing so I can have some feedback.

oh, ok, so what do you mean when you say beta?

this is very educational

[Lev]

Half way between Alpha and Final

[frodoontop]

When I say "test" I really mean testing so I can have some feedback.

What kind of feedback exactly? On problem samples, normal music? Which bitrate (vbr/cbr) should have improvements, if any? What was the overall goal of 3.97?

[Gabriel]

oh, ok, so what do you mean when you say beta?

Beta should be usable for everyday use, but are usually less tested for stability, compilation across different platforms , ... than releases.

What kind of feedback exactly? On problem samples, normal music? Which bitrate (vbr/cbr) should have improvements, if any? What was the overall goal of 3.97?

NEXT (not yet available) alpha. I'll probably ask for some feedback when it will be available.

[Jojo]

is there any changelog available?

[Gabriel]

not yet

[mithrandir]

The most recent builds of alpha4 have an annoying bug. For instance, if you add "--shortthreshold 4.5,35" to the commandline (e.g. -V 5 --shortthreshold 4.5,35) the output is poor quality (mushiness and swishiness in transients) even though this particular setting is more stringent than the V5 default. Therefore, something is broken.

Note that an alpha4 build from November 2004 (which I used to have) didn't have this problem.

This post has been edited by mithrandir: Jan 8 2005, 08:02

[Gabriel]

For instance, if you add "--shortthreshold 4.5,35" to the commandline

And if you don't add it?

[Gabriel]

3.97a5 is comitted in cvs.

I'd mainly like opinions about "-b 128 -X 10,10" against 3.90.3.

It also includes a few tuning changes to all presets.

[Enig123]

Can someone tell me how to cvs this alpha branch? What's the cvs command exactly?

[mithrandir]

For instance, if you add "--shortthreshold 4.5,35" to the commandline

And if you don't add it?

If you don't add it, the output sounds fine.

[[proxima]]

3.97a5 is comitted in cvs.

I'd mainly like opinions about "-b 128 -X 10,10" against 3.90.3.

It also includes a few tuning changes to all presets.

I've just finished a quick test:

CODE

              3.90.3_ap cbr 128   -b 128 -X 10,10    3.961_V5   3.97a5_V5
applaud.wav             2.0           1.5                3.2         2.9
Bayle - Etching.wav     1.8           1.0                2.2         2.5
BigYellow.wav           2.3           3.0                5.0         5.0
chanchan.wav            4.0           3.5                3.2         2.8
FloorEssence            2.5           2.0                2.5         2.5
Mahler.wav              3.8           1.5                3.3         2.0

IMHO the new cbr settings still suffers of the infamous ringing (Bayle - Etching.wav) and warbling (applaud), sometimes preecho is worse than 3.90.3. I can provide samples and ABC/HR logs on request.

This post has been edited by [proxima]: Jan 9 2005, 19:23

[sven_Bent]

stable doesn mean not crashing
stable mean not making wierd results on some samples

[amano]

*amano hugs proxima*

for the time consuming ABXing!

[LoFiYo]

I quickly tested the new alpha with hihat.wav. 3.90.3's --alt-preset cbr 128 still handled this particular sample much better. The first hihat hit is heavily distorted with 3.97a5. BUT this sample may be considered invalid according to Gabriel's proposal. I will try other samples later.

[LoFiYo]

I used fatboy this time.

3.97a5 (-b 128 -X 10,10) VS 3.90.3 (--alt-preset cbr 128)


ABC/HR Version 0.9b, 30 August 2002
Testname: fatboy 3.90.3 VS 3.97a5

1R = C:\My Test Samples\fatboy\3.90.3.mp3.wav
2R = C:\My Test Samples\fatboy\3.97a5.mp3.wav

---------------------------------------
General Comments:
#1 sounds better to me. #2 sounds really "dirty" and coarse.
---------------------------------------
1R File: C:\My Test Samples\fatboy\3.90.3.mp3.wav
1R Rating: 1.5
1R Comment: lots of noise is added, but still sounds cleaner and closer to the original overall.
---------------------------------------
2R File: C:\My Test Samples\fatboy\3.97a5.mp3.wav
2R Rating: 1.0
2R Comment: less noise is added, but distorion is too severe and sounds dirty.
---------------------------------------
ABX Results:
Original vs C:\My Test Samples\fatboy\3.90.3.mp3.wav
7 out of 7, pval = 0.008
Original vs C:\My Test Samples\fatboy\3.97a5.mp3.wav
7 out of 7, pval = 0.008
C:\My Test Samples\fatboy\3.90.3.mp3.wav vs C:\My Test Samples\fatboy\3.97a5.mp3.wav
7 out of 8, pval = 0.035

[dev0]

Could anybody test this sample with the new LAME alpha:
http://dev0.rc55.com/files/samples/giveupt...ample18sec.flac

It caused serious problems (similiar to HiHat.wav) with 3.96.1 and I have to leave for school right now.
All alpha5 testing should go into this thread from now on for the sake of clarity.

My own tests will follow later.