Yesterday, in another forum (no, not r3mix.net )...

nazz wrote:



nazz courteously provided a sample. I uploaded it (LPAC compressed) to www.citay.de/links/sumo/dogwhistle.pac and made a small listening test with it (on AKG K290's):

"--alt-preset 128": Torture. The high frequencies seem to be missing altogether. It's not a whistle anymore. I soon deleted this one, because it totally disturbed my feeling on how the wistle should sound.

"--alt-preset standard": Has big problems with it... chirping nervously. Easy to discern.

"--alt-preset extreme": Same as standard, chirping, warbling, however you call it.

"--alt-preset fast standard": Doing quite alright, for my ears.

"--r3mix": Also doing okay. Slightly worse than fast standard, i would say.

"--alt-preset insane": Sounds flawless to me. I don't know how nazz could ever distinguish this from the original, heh..


somebody (yes, that's a nickname) also confirmed these findings. I know this may fall under the "artificial sample" category... but hey, it's from "A Day In The Life" (The Beatles - Sgt. Pepper's Lonely Hearts Club Band, Track 13).

Insane sounded the closest to me.

But can any dogs tell the difference?

haw haw

Hi. I encoded this cd awhile back with 3.88. Commands were --abr 192 -h -ms --lowpass 19.5 and maybe -Z.

It sounds very much like the original LPAC. No gross artifact.

I won't be able to check this out for awhile since I just moved and don't have much setup to listen on.

The problem is most likely due to an issue somehwere with noise shaping 2, but that's odd because standard doesn't have the problems fast standard did with it in most situations, and even when using noise shaping 2, it still usually comes out sounding far better than --r3mix. If this problem is more apparent because of noise shaping 2, it's probably due to a flaw in the masking used.

There are enough of these "problems" that it's very apparent that LAME needs a massive overhaul in this department, but I'm not sure who, if anyone, is going to do it.

You can try adding -Z if you like.. that may fix the sample for --alt-preset standard, but I certainly wouldn't recommend doing this in all cases.

Could it be a problem with --vbr-old which isn't in --vbr-mtrh?
I mean both --r3mix and --alt-preset fast uses --vbr-mtrh while the problem presets use --vbr-old

I doubt it. Both --r3mix and --alt-preset fast also completely disable noise shaping 2, which I believe is the cause of the problem. Apparently even my adaptive mode, which seems to work in every other scenario I've seen yet (aside this one... though I haven't tested this sample yet either), including fatboy and all the impulse cases in addition to working fine on low volume vocal clips, doesn't quite cut it here. This is the reason why I still use an adaptive noise shaping method which switches between type 1 and 2 with vbr-old.. because in nearly all cases, it seems to work fine (this is the first I've seen where it doesnt), but vbr-mtrh still has big problems in some cases with type 2... which is why I disabled it in the "fast standard" mode.

To further comment on this situation: this one clip aside, the average still shows with overwhelming evidence that --alt-preset standard with an adaptive noise shaping method and vbr-old outperforms --r3mix (with mtrh and no type 2 noise shaping at all) in very many situations.