psy.c: CalculateSMR ?

andym00 · 13 May 2007, 09:07 am

Just curious, but I've been following Musepack since it's original invocation, especially it's psychoacoustics and have always been a massive fan its results, and I've just noticed that the signal to mask ratio calculations have changed (significantly) since I last looked at it..

Can one of the developers explain to me why during signal to mask ratio calculations that the peak SMR discovered in a FFT resolution thresholds when find for each subband are now scaled by 1.44121959671885364405f ?
Previously, as far as I can see, it never used to inflate the SMRs like that, merely scaling by 0.0625f and which this additional scaling the bitrates seem overinflated.. Over inflated for a given NMT & TMN.. Maybe as much as 20%-30% increase in final bitrates..

Has in previous previous incarnations the psychoacoustics been miscalculated ? Or am I missing something here ?

I can't actually see when this change was made, but am very curious about it..

Lefungus · 13 May 2007, 08:06 pm

Can you post more details ?
I only see that in psy.c:835 which seems to contradict what you're saying.

Code:

        *smr0++ = 0.0625f * *erg0++ / tmp0;

In any case, that code was like that in the first public tarball of mppenc code from Klemm

andym00 · 14 May 2007, 11:00 am

Ahh, okay, I see now..
In SVN it's as its always been, but in the musepack-1.95z67.zip I was looking at it's changed..

13 May 2007, 09:07 am	#1
andym00 Guest Join Date: Jan 2006 Posts: 3	psy.c: CalculateSMR ? Just curious, but I've been following Musepack since it's original invocation, especially it's psychoacoustics and have always been a massive fan its results, and I've just noticed that the signal to mask ratio calculations have changed (significantly) since I last looked at it.. Can one of the developers explain to me why during signal to mask ratio calculations that the peak SMR discovered in a FFT resolution thresholds when find for each subband are now scaled by 1.44121959671885364405f ? Previously, as far as I can see, it never used to inflate the SMRs like that, merely scaling by 0.0625f and which this additional scaling the bitrates seem overinflated.. Over inflated for a given NMT & TMN.. Maybe as much as 20%-30% increase in final bitrates.. Has in previous previous incarnations the psychoacoustics been miscalculated ? Or am I missing something here ? I can't actually see when this change was made, but am very curious about it..

13 May 2007, 08:06 pm	#2
Lefungus Procrastinator Join Date: Jul 2004 Posts: 131	Can you post more details ? I only see that in psy.c:835 which seems to contradict what you're saying. Code: smr0++ = 0.0625f *erg0++ / tmp0; In any case, that code was like that in the first public tarball of mppenc code from Klemm

Thread Tools	Search this Thread
Show Printable Version Email this Page	Search this Thread: Advanced Search
Display Modes
Linear Mode Switch to Hybrid Mode Switch to Threaded Mode

14 May 2007, 11:00 am	#3
andym00 Guest Join Date: Jan 2006 Posts: 3	Ahh, okay, I see now.. In SVN it's as its always been, but in the musepack-1.95z67.zip I was looking at it's changed..