Should I upgrade to LAME v3.99? Options

[SonicBooom!]

Hello there! Quite a while since I last visited this forum. Seems like LAME has released a new version which is v3.99. I would like to ask if should I upgrade to this new version. I'm currently using v3.98.4 as the encoder and -V 2 as the preset when converting my FLAC/CDs to MP3s. Is there any audible differences/regression to 3.99 compared 3.98.4? Any help for the non-techie will do. Thanks

[Canar]

I've switched. I haven't noticed any regressions. It's been perfectly transparent in all the applications I've used it in. I haven't been trying to break it though or really testing hard or anything.

[SonicBooom!]

I encoded Song X to 3.98.4 and 3.99.3 and here's the difference I found:

CODE

Tool :        LAME3.98r                              LAME3.99r
File size :  6.75MB (7081929 bytes)          5.76MB (6041127 bytes)
Bitrate :    228 kbps                        201 kbps

Is the difference in file size and bit rate has something to do with quality?

[lvqcl]

Not necessarily.

[SonicBooom!]

So is this a good thing that the file size has lowered or is it a bad thing that the bit rate lowered? Why is 3.99 like this?

[onkl]

File size and bitrate are essentially the same. The encoder decided it needs fewer bits then before, which could mean it's more efficient or it underestimated the difficulty. Only blindtesting for audible differences will give a definite answer.

[Brand]

With one sample I tried (this one here), 3.99.3 actually produced a larger and I think also slightly better sounding [more transparent] file than 3.98.4.
So I guess it really depends on the material.

[psycho]

Since 3.99 became 3.99.3 I recommend it. The smaller filesize you're noticing is probably because of extensive VBR optimization that happened in 3.99. The -V 0 was particularly what the LAME devs worked on, but I would imagine this also had some effect on other -V switches.
Until now, I have not yet encountered a sample, where 3.99.3 would be worse than 3.98.4. Also -V 0 should produce bigger filesizes with 3.99.x compared to previous LAME version, however that is probably true for average music. Since I encode almost entirely metal I haven't noticed this supposed filesize change - it probably has something to do with high frequency bloating. I imagine that with versions before 3.99.x, I got big mp3 files because of this problem and now with 3.99.x I don't see a filesize difference, because with -V 0 that I use, now there is no lowpass. My logic suggest that now -V 0 uses those bits better than the versions before and that I actually benefit from it with 3.99.x, where I probably was wasting space before.

Also, with some songs I have noticed both bigger and smaller mp3 filesizes with the new version.

[SonicBooom!]

psycho, thanks! The word "optimization" is just what I've been waiting to hear My brother though noticed some kind of pre-echo on the new version when we encoded the same track on 3.99.3 and 3.98.4, but it's very subtle and I rarely listen that attentively so it might not be a problem.

Thanks for the help, guys! I'm using 3.99.3 now

[apodtele]

I encoded Song X to 3.98.4 and 3.99.3 and here's the difference I found:

CODE

Tool :        LAME3.98r                              LAME3.99r
File size :  6.75MB (7081929 bytes)          5.76MB (6041127 bytes)
Bitrate :    228 kbps                        201 kbps

Is the difference in file size and bit rate has something to do with quality?

High end VBR target rates has indeed changed with -V1 and -V0. They stayed close to the original for -V2 and below.
The bitrate is very much the quality measure for a mature encoder like LAME.

[d_headshot]

High end VBR target rates has indeed changed with -V1 and -V0. They stayed close to the original for -V2 and below.
The bitrate is very much the quality measure for a mature encoder like LAME.

So I should stick with v3.98 rather than upgrading?

[Canar]

So I should stick with v3.98 rather than upgrading?

Unless you can provide ABX tests proving regression, I'd say that there's no reason not to upgrade.

[greynol]

Are we upgrading to save space? If we think we can get the same sound quality out of a smaller file then TOS #8 applies.

Are we upgrading to improve quality? If we think we can get better sound quality for the same size file then TOS #8 applies.

Are we upgrading to get bigger VBR files than what was possible from the previous version? If we're upgrading based on audible quality differences or some paranoid-based phenomena such as "margin" or "defensiveness", then TOS #8 applies.

Personally, I think upgrading for the sake of there being a new version is rash.

This post has been edited by greynol: Dec 20 2011, 00:43

[Takla]

Are we upgrading to save space? If we think we can get the same sound quality out of a smaller file then TOS #8 applies.

Surely TOS#8 refers to demonstrating a difference, not proving non-difference? For example, from
http://www.hydrogenaudio.org/forums/index....st&p=777840:

A critical point to mention is that this forum does not and never will require someone to provide evidence that two things sound the same.

Which seems succinct, rational, and precise.

So if person A states that they went from version X to version Y because the file size is smaller and he/she says they can't hear any difference then asserting TOS#8 looks unwarranted.

[Canar]

Personally, I think upgrading for the sake of there being a new version is rash.

If that's the sole reason, I agree with you, but banging on new LAME versions is valuable to the community. The only way for it to be tested is for people to use it. I'd say that the community's position should be to recommend the latest and greatest until regression is proven, if for no other reason than finding regressions.

[greynol]

So if person A states that they went from version X to version Y because the file size is smaller and he/she says they can't hear any difference then asserting TOS#8 looks unwarranted.

Yes, with the specific case where there is transparency with both versions you're absolutely correct.

Now suppose someone is using something like -V5 and at least one version of Lame is not transparent. In this case a double blind test would be warranted.

@Canar, I appreciate your point of view. This is a bit different than the insanity of converting everything to the newest version of flac, wasting electricity and putting additional wear on hard drives in order to save a tiny fraction in storage. I suppose one could make the case that we should encourage this in order to flush out bugs, or maybe not if that means that once perfectly good data has now become corrupted. Anyway, I digress; I do think building confidence in the newest version of Lame is a good thing.

This post has been edited by greynol: Dec 20 2011, 05:11

[d_headshot]

I upgraded to it anyways since my family got a new computer and I had to install some programs again. I noticed that when converting to v3.99 from flac, it takes foobar2000 quite a long time compared to 3.98. I don't think it's a problem with foobar or my computer because I'm running on a 3.3GHz processor(and converting with a 2GHz processor was faster). I really don't know what's going on lol.

[db1989]

I seem to recall there being a fair bit of discussion about speeds, compilers, optimisations, etc. in the news thread; you may find some explanation there.

[goa pride]

Hello there! Quite a while since I last visited this forum. Seems like LAME has released a new version which is v3.99. I would like to ask if should I upgrade to this new version. I'm currently using v3.98.4 as the encoder and -V 2 as the preset when converting my FLAC/CDs to MP3s. Is there any audible differences/regression to 3.99 compared 3.98.4? Any help for the non-techie will do. Thanks

upgrading increases the size and quality of mp3 vbr -V 0 from ~256 kbps to ~270 and more (this happen with powerful electronic music)

[lvqcl]

There's no direct relationship between size and quality.

[pdq]

There's no direct relationship between size and quality.

But all else being equal, wouldn't you say that having more bits to work with generally allows more accurate reproduction?

[JJZolx]

I switched to LAME 3.99.3 just for its ability to add TXXX fields and embed images larger than 128k. My Mp3 library consists entirely of transcoded FLAC files, so these changes make transcoding easier by requiring fewer 3rd party tools to deal with the tagging limitations of previous versions. Comparing the sizes of the libraries encoded at -V2 with both 3.98.4 and 3.99.3, the total size grew by about 3% with the new version.

[FreaqyFrequency]

But all else being equal, wouldn't you say that having more bits to work with generally allows more accurate reproduction?

Absolute accuracy isn't really the end goal with a lossy codec, but rather audible transparency to a human being. If a human being is unable to distinguish between a 150kbps encoding and a lossless original in a controlled, blind listening test, then an argument that the file needs more bits to be "more transparent" doesn't bear out. A 450kbps encoding would do no better in that test if it is already audibly transparent at 150.

This post has been edited by FreaqyFrequency: Jan 16 2012, 02:27

[Satellite_6]

Surely there is a direct relationship between size and quality in an absolute sense. . .

[saratoga]

Surely there is a direct relationship between size and quality in an absolute sense. . .

I'd say there is an indirect relationship.