Can anyone explain to me how the 18.6ms are calculated in http://wiki.hydrogenaudio.org/index.php?ti...hase_filterbank with regards to the impulse response of the short blocks?
Even more funny:
"At 48kHz sampling rate a long block has a time resolution of about x ms."
???
"At 48kHz sampling rate a long block has a time resolution of about x ms."
???
QUOTE (Sebastian Mares @ Jun 2 2009, 21:50) 
Can anyone explain to me how the 18.6ms are calculated in http://wiki.hydrogenaudio.org/index.php?ti...hase_filterbank with regards to the impulse response of the short blocks?
The bandpass filters of the 32 subband filterbank have a response of 512 samples. The impulse responses of the filters in the MDCT filterbank for short blocks hace an impulse response of 12 but this is at 1/32 of the sampling rate. Now, to get the length of the combined impulse responses: 512 + 32*12 - 1 = 895 samples. At 48000 Hz this corresponds to 18.646 ms. OTOH most of samples (probably the beginning and ending 5 ms -- I havn't checked) are close to zero. So, it's actually not that big a deal.
Cheers!
SG
This is a "lo-fi" version of our main content. To view the full version with more information, formatting and images, please click here.