Yet another VBR killer Sample (FHG VBR does well though) Options

[Polouess]

this is the start of a movie trailer I had to encode for the web, so the target was not transparency but "not annoying". as target bitrate, considering bandwith and the video bitrate, I chose 144 kbps (first I wanted to go even lower but this special sample got really annoying then).

I uploaded only the first few secs, the whole thing can be listened to at www.thesoundofindia.de

LAME 3.98b8's VBR totally fails here, it produces annoying distortion.
ABR does quite well at the same bitrate, and FHG in VBR mode (which I suppose is similar to LAME ABR) does slightly better.

the differences are even bigger in real life, that is when I encode the whole track and not only these few seconds, as different settings are needed to produce the target bitrate.

the original sample

this sample at 144k:
FHG (tho one shipping with Audition 1.5) VBR 52
LAME 3.98b8 --preset 153 (do yourself)
LAME 3.98b8 -V 3.4 (do yourself)

the same sample, but with the settings I used to make the whole track 144k:
FHG (tho one shipping with Audition 1.5) VBR 63
LAME 3.98b8 --preset 144 (do yourself)
LAME 3.98b8 -V 4.55 (do yourself)
in this case, which is my "real-world example", LAME is much worse, as you'd expect.

for curiosity, I tested 3.97final against 3.98b8 at -V2. there is no improvement in my opinion, especially when you consider the bigger file b8 produces. not sure about that though, we have noisy craftsmen in the house

[shadowking]

Is it artificial ? It sounds horrid but i think old vbr is a bit 'better' here. MPC Q5 sounds fine ~230k

[Polouess]

Is it artificial ? It sounds horrid but i think old vbr is a bit 'better' here. MPC Q5 sounds fine ~230k

Not really artificial.
It's a North Indian instrument called Tanpura, which has an extremely rich overtone spectrum. However, it was recorded with a camera mic in a bad room, so some post processing (pseudo stereo, EQ) was applied.