Skip to main content

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: Preliminary Results from Non-Linear Psymodel Patch... (Read 4374 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Preliminary Results from Non-Linear Psymodel Patch...

Yesterday I downloaded and experimented with Alexander's new patch for the non-linear psymodel in LAME.  Instead of typing all that I've said about it again, I'll just quote the conversation off of irc:

Quote
<Dibrom> seems to help gpsycho alot on clips with tonality problems like 2nd_vent and serioustrouble
<Dibrom> but overall, nspsytune with -X1 is sounding better
<Dibrom> except for on fatboy
<Filburt>
<Dibrom> but on fatboy nspsytune plus -X3 sounds better *without* the new psymodel than with it
<Dibrom> also seems to bloat the bitrate pretty massively on certain clips
<Dibrom> I'm going to try and tweak some of the values a little later this week and see what I come up with
<FastForward> what does -X1 do better on, and what does -X3 do better on?
<Dibrom> nspsytune + x1 does way better on 2nd_vent than the new model with gpsycho.. the new model eliminates the ringing, but the pre-echo is still there, both are gone with nspsytune + x1
<Dibrom> nspsytune + x3 does even better on all of those
<Dibrom> now it gets kind of confusing
<Dibrom> on fatboy
<Dibrom> --dm-preset standard as it is now (nspsytune + x1) sounds better with the new psymodel
<Dibrom> but when you use +x3 it sounds better still, and if you use the new psymodel WITH that, it sounds worse
<Dibrom> so
<Dibrom> the non-linear model usually seems to be a little worse or equal to nspsytune + x1
<Dibrom> except on fatboy
<Dibrom> if you use the non-linear model with nspsytune and +x1 it seems to be pretty much the same, just the bitrate increases alot
<FastForward> interesting
<Dibrom> there is one particular value that you can tweak which apparently when modified to 1 should equal the old psymodel, right now its set at .3
<Dibrom> I'm going to see at which point it seems to eliminate most problems with gpsycho but without increasing bitrate too much with nspsytune + x1
<Dibrom> and maybe even see if I can find a threshold where it doesnt sound worse with + x3


So for a quick summary, with gpsycho this patch seems to help, but with nspsytune (which defaults -X1, except for with --r3mix) this patch doesn't do much except increase bitrate.  This may not be so surprising considering that in most cases, nspsytune with maximum noise measuring will eliminate many of the same problems I believe this patch is designed to do.  The only advantage the patch seems to have over nspsytune at the moment is on impulses and only when -X1 (default) is used.  With -X3, nspsytune outperforms this patch even on impulses, and actually using the patch in that situation decreases quality slightly, or rather doesn't allow it to go quite as high.

Anyway, all of this is very preliminary... and all based on an alpha quality patch that right now probably only implements about 25% of the entire idea for this new model and currently this is only some initial testing on some of the worst problem clips I know of for LAME.  I still have some pretty high hopes for this overall.. and of course I appreciate the work it has taken to start to get this implemented, but obviously (and of course nobody isn't aware of this) there is still some tuning required.  Right now, I'm still thinking that improved noise measuring may actually help a little more overall.

Btw, those interested in testing this can download it from http://mitiok.free.fr (courtesy of mitiok, thanks for the compiles )

Please remember though that this is only a very experimental patch at the moment, not meant for common use.  Please don't complain about the quality if you find problems with it, but posting your findings here or letting Alexander know on the dev list would be welcome I'm sure.

Preliminary Results from Non-Linear Psymodel Patch...

Reply #1
a bit off topic but....

i see filburt!

where was this chat from?

 

Preliminary Results from Non-Linear Psymodel Patch...

Reply #2
#project_mayhem on irc.openprojects.net of course  Accessible from the chat button up top also.