Verify waste sample (Lame cbr 128) Options

[Schlotter]

Could anybody verify my listening test? I can hear horrible artifacts with Lame 3.98b3 and Lame 3.97 --preset cbr 128. At 192kbps its ok. What is this around the guitar-echo? Lame 3.96 sounds better with --preset cbr 128.

Attached File(s)

 waste_original.flac ( 1.83MB ) Number of downloads: 185

 

[shadowking]

Confirmed nasty warbling on 128~160 cbr/abr. At 128~160, it is not uncommon for different LAME versions to do different things. High quality settings (V2 +) are more secure with all the versions. Differences are only on killer samples. At lower bitrates there is a phenom of artifacts appearing on one version but not the other / on new vbr and not old vbr and vice versa. I also believe you can get reasonable security with cbr/abr no matter what the encoder version is - but only at high bitrates (192 +).

[Schlotter]

I personally prefer -V 2 --vbr-new. I just don't know what make this sample so hard to encode proberly. Only Lame 3.98b3 and 3.97 sound crummy on this with --preset cbr 128, far worse than 3.88 with -b 128 -h !?

This post has been edited by Schlotter: Jun 9 2007, 15:51

[shadowking]

3.88 used a different psymodel. Its possible that an inferior encoder does better on the odd sample. Generally cbr 128 is a bad idea for quality -v5 is more interesting.

[Schlotter]

Do you think waste_original is a killer sample? I can hear the difference --> original --> preset insane (ABX15/15) . Preset standard work fine on this sample with all versions, but not perfect.

Another one: mama_original.flac (just guitar strumming)

Lame 3.98b3 & 3.97:
preset cbr 128 --> worst, described as "warbling"
-V 5 --> less warbling, audibly there, no alternative !

Lame 3.96:
preset cbr 128 --> better, slight warbling
-V 5 --> same

Lame 3.88:
-128 -q 2 --> best in my opinion

If I had to choose a version, it would probably be 3.96 with -V 5 or 3.88 with -b 128 -q 2 for mp3 at low bitrates ~ 128kbps.

Addendum: Second test I used Sony CD-MDR 370 headphones, at night. 3.96 with V 5 --> slightly more worse than --preset cbr 128.

This post has been edited by Schlotter: Jun 20 2007, 15:29

Attached File(s)

 mama_original.flac ( 1.08MB ) Number of downloads: 211

 

[Kjn_Wds]

I personally prefer -V 2 --vbr-new. I just don't know what make this sample so hard to encode proberly. Only Lame 3.98b3 and 3.97 sound crummy on this with --preset cbr 128, far worse than 3.88 with -b 128 -h !?

can anybody please explain me what -b## (and maybe also -mj) switches are \ howdo they work work :°° ?

[Kjn_Wds]

I personally prefer -V 2 --vbr-new. I just don't know what make this sample so hard to encode proberly. Only Lame 3.98b3 and 3.97 sound crummy on this with --preset cbr 128, far worse than 3.88 with -b 128 -h !?

can anybody please explain me what -b## (and maybe also -mj) switches are \ howdo they work work :°° ?

Solved, btw

[Schlotter]

Second test: 3.88 (-b 128 q2), 3.90, 3.91, 3.92 (--alt-preset 128, --alt-preset cbr 128, -V 6), 3.93.1, 3.96, 3.97, 3.98b3 (--preset cbr 128, --preset 128, -V 5).

Only 3.97 and 3.98b3 seems to have the problem with this heavy warbling at bitrates around 128kbps on both samples.

Add
ABC/HR CBR 128 Results: mama_cbr128_test
ABC/HR V 6/V 5 Results: mama_V_test
ABC/HR ABR 128 Results: mama_abr128_test

This post has been edited by Schlotter: Jun 23 2007, 16:05