here is a proposal for a LAME lofi quality vbr setting that should result in 140 kbps on average (+/- 40):

-mj --vbr-new -V5 -q1 --athtype 0 -Z -X4 --cwlimit 22 -k --athlower -8 --athaa-type 2

comments?

Of course, for those trying out this line, the natural thing would be to compare with --alt-preset 140

of course, --alt-preset 140 is not a real vbr mode

I was going to suggest the same thing. There needs to be some sort of a baseline to compare against, even if the modes aren't strictly the same. There's bound to be artifacting of some sort. The question is whether the quality is better or worse than the baseline, on average.

I would also suggest that the bitrate estimate be verified, perhaps by encoding a couple different albums.

ff123

regarding the bitrates, it seems to have a similar behavior than the vbr-presets.

[img]dac2b43349[/img]

Some listening results here:

http://ff123.net/export/robert.html

The worst performance was on rach_original.wav, where the bitrate dipped down very low and it sounded very bad.

ff123

This is well suited for portables :listening

mpropper, it would be more interesting if you compared against --alt-preset 140, which is abr (--alt-preset-standard / extreme are vbr)

i think the intention is to find a good vbr-setting "below" --alt-preset standard (e.g. for portable players).

Yes, but the first test it has to pass is being better than --alt-preset 140. If it's not better than that, then there really isn't any point using it over --alt-preset 140.

From the results reported by Robert, that vbr setting doesn't look too promising. It's better than --alt-preset 140 when the resulting bitrate is much higher, and worse than --alt-preset 140 when the bitrate dips, and similar results when the bitrate is around 140kbps. Doesn't help that the bitrate fluctuates a lot, almost randomly, so the result you get will be pretty much random. It doesn't achieve VBR's intention to keep quality consistent across samples.

I think I'd wait for for Dibrom to start tuning ABR with the custom stuffs he has been using for the --alt-preset standard/extreme, or to tune lower bitrate VBRs.

Or perhaps the settings suggested can be improved, maybe add in --nspsytune --ns-bass -4 ?

I think the problem, which Dibrom has alluded to before, is that LAME's psychoacoustic model used for VBR really isn't that great. It works fine at higher bitrates, but when you dip to lower bitrates, it starts being almost random and stops consistently distinguishing between "audible" and "inaudible" noise. To do a reliable low-bitrate VBR would probably require a new psychoacoustic model specifically tuned for low bitrates (like the way FhG does it).

So, any conclusions from your side yet, Robert?