Just another thought:
I believe that the concerns regarding the SRC performance as outlined at ht=
tp://src.infinitewave.ca are a bit academic with little practical consequen=
ces at least for typical nature recording.
Note that they used an incredibly loud -6 dbFS synthetic sine sweep, spanni=
ng the frequency range from 0 to 48 kHz. However, a typical nature recordis=
t will probably never face such strong ultrasonic signals (due to both the =
limited frequency responses of the microphones and the strong atmospheric a=
ttenuation at those frequencies). In addition to that, the inherent microph=
one noise floor will mask most of the low-level aliasing products that are =
visible there.
Consequently, the demands for the low-pass filter in a 88.2 to 44.1 ks/s SR=
C algorithm are quite moderate because there is almost no energy above abou=
t 20 kHz that could be folded-down into the audible range.
On the other hand, an upsampling algorithms would introduce relatively stro=
ng ultrasonic components that were not present in the original recording (t=
hey just insert zeros between the original samples!!!). For this reason, th=
ey require a more demanding low-pass filter with a better stop-band attenua=
tion (which is requires more computing power).
Regards,
Raimund
|