SRC Comparisons (96 kHz to 44.1 kHz)
Jan 22, 2006 at 12:33 AM Post #2 of 6
OK, now where do I get (or how do I generate) a good enough sample of a 1 kHz tone in 96 kHz 24 bit... what Audacity generates apparently isn't cutting it.
 
Jan 22, 2006 at 1:17 PM Post #3 of 6
Quote:

Originally Posted by sgrossklass
OK, now where do I get (or how do I generate) a good enough sample of a 1 kHz tone in 96 kHz 24 bit... what Audacity generates apparently isn't cutting it.


watch in foobar @ Playback -> Input -> Standard Inputs and use the Diskwriter.
 
Jan 22, 2006 at 3:50 PM Post #4 of 6
Seems it wasn't Audacity, but the spectrum analyzer settings. When using RMAA for analysis, it seems best to employ a Kaiser window of 20 along with the largest FFT size. The resampling quality for 96k -> 44.1k 24 bit came out like this: SSRC (high precision exe, twopass) > SRC (libresample via SRCdrop) >> Audacity.
 
Mar 9, 2006 at 8:09 PM Post #6 of 6
Well, I don't have Matlab here, but I've played with a 20 kHz tone generated in 24/96 and resampling to 44.1 kHz.
As before, SSRC (ssrc_hp.exe, twopass) yielded the best result, with only 3 spurious tones visible above the noise level at -180 dBFS, all of them under -160 dBFS.
SRCdrop (with "best" resampler) gave a "livelier" spectrum, though only three tones reached above -150 dBFS, with one of them being direct aliasing at about -110 dBFS and the only one above -140 dBFS. (In practice, you can assume a background noise floor at about -140 dBFS even with the best of DACs.)
And again, Audacity 1.2.4 with its "High Quality Sinc Interpolation" yielded the worst results by far, with a whole zoo of spurious tones, of which about 11 reach -110 dBFS, 4 are above -100 dBFS and one (again, direct aliasing from 20 kHz) reaches -84 dBFS. Don't resample here!
 

Users who are viewing this thread

Back
Top