Problem samples for Lame

2005-08-31 09:43:02

[attachment=1719:attachment]

[attachment=1718:attachment]


The first sample seems to be a solo accordion or it may be even electronic. I was testing gogo 3.13 and was sure it was the encoder stuffing up, but recent Lame 3.97 is not better. Quality at 128k is terrible, at V2 vbr-new quality is also poor - abx 8/8, at V0 vbr-new 12/16, maybe a bit better.

Basically it doesn't sound right at all. The original is an accordion and the mp3 sounds like an toys r us accordion under a shower.



The next sample is terrible at 128 abr, 160 abr, V4 vbr new, V3 vbr new is even worse and V2 vbr new -Y is also not good. I am also suspecting that normal V2 vbr-new  is also not transparent but didn't have time to test. Vbr-old is better but still not that good, vbr-new the same as abr. The loudest artifact is at 2 secs position, but around 19 secs is also effected. Outside of abx the whole thing sounds dodgy at 128k abr, but V3 also doesn't sound that good at least not for the first 2 seconds.

Problem samples for Lame

Reply #1 – 2005-08-31 10:02:29

which LAME?

Problem samples for Lame

Reply #2 – 2005-08-31 10:50:52

3.97a11 and 12.

3.90  and gogo are better on 'richmond rd'. I didn't test the accordion on earlier versions but suspect it will be worse due to ringing.

Problem samples for Lame

Reply #3 – 2005-08-31 11:05:19

gogo uses GPSYCHO psy model without any changes by Naoki. What commandline arguments did you use for 3.90 ?

Problem samples for Lame

Reply #4 – 2005-08-31 11:11:09

For 3.90 I used --preset 160 or 128

Problem samples for Lame

Reply #5 – 2005-11-09 13:49:02

Conducted a few ABX tests today with the Karov_El_Libech2 sample: 128 kbps (CBR, ABR and VBR) and --preset medium (regular and fast).