LAME VBR CBR Confusion Options

[whootz0r3d]

Been reading this article about LAME http://wiki.hydrogenaudio.org/index.php?title=LAME but I'm still confused about wheather VBR V0 / CBR 320 which one is better than the other? Some paragraphs in that article seem to contradict one another..

Without getting into too much details which one IS technically the superior and why in plain english?

This post has been edited by whootz0r3d: Nov 17 2012, 14:49

[[JAZ]]

The more you look into a definitive answer on that, the more you'll be reducing the answer to fewer cases.

The simple answer is that they try to do the same, while V0 is able to reduce the bitrate requirements.

In LAME 3.99, the internal tunings where changed so that V0 is stronger, and there is also halb97's modified lame, which pushes this further.

On the other hand, CBR in Lame 3.99 is also different to what it was in, for example, Lame 3.96. It was seen that applying the VBR algorithm with a fixed bitrate improved on some samples.

As such, there are many differences between versions on what "V0" is and what CBR320 is. The consequence of this is my initial sentence. You can't get a definitive answer so it all ends up on if you want or don't want the benefits of VBR, or if you still have a (rare) player that works worse with VBR files.

[soundping]

My preferred MP3 setting: VBR -V0 -q 0

VBR has an advantage of adjusting to the finer elements of sound. Music with string instrument sounds for example.

CBR will not adjust and the finer elements could be lost or distort.

[Dynamic]

I think the consensus at such bitrates is that problem samples are very rare for either mode and are usually fairly subtle and not too unmusical in nature.

halb27's modified lame in -V0+ mode fixes a certain group of fairly rare problem samples (particularly short block tonal samples) and improves a few more without making them completely transparent. It tries to reserve as much bit reservoir as possible to allow momentary bitrates to greatly exceed 320kbps (e.g. 480kbps or so) for key frames in many of these samples, where a constant 320kbps bitrate approach might not reserve as much bit reservoir.

I'd be tempted to recommend -V0+ in halb27's modified lame as about as robust as you'll get with MP3 at the moment if you don't care about file size, though there may be future tunings that improve it further, especially given internal psymodel improvements in lame 3.100 alpha2. Proving or falsifying that when problem samples are so rare, however, is next to impossible.

[nastea]

My preferred MP3 setting: VBR -V0 -q 0

As far as I know the -q 0 option will not make any difference since -V0 already uses the highest quality algorithm.