Next |
Prev |
Up |
Top
|
JOS Index |
JOS Pubs |
JOS Home |
Search
Do we need FFT convolution for practical audio filters?
Yes:
- FFT convolution [
lg
] starts beating time-domain
convolution [
] for
or so (on a single CPU)
- The nominal ``integration time'' of the ear, defined, e.g., as
the reciprocal of a Bark critical-bandwidth of hearing, is greater
than 10ms below 500 Hz
- At a 50 kHz sampling rate, this is 500 samples
- FIR filters shorter than the ear's ``integration time'' can
generally be characterized by their magnitude frequency response (no
perceivable ``delay effects'')
- Thus, even ``perceptually instantaneous'' FIR filters can easily
be hundreds of taps long
- For longer FIR filters, the FFT advantage is that much greater
- We conclude that FFT convolution is an important implementation
tool for FIR filters in digital audio
Next |
Prev |
Up |
Top
|
JOS Index |
JOS Pubs |
JOS Home |
Search
[Comment on this page via email]