Frequency response of the fractional delay component

Hello larmac,

Yes, it's the fractional delay's normal response. The fractional delay works by splitting the difference between the two adjacent whole-numbered delays. Given a requested delay of N.x, where N is an integer and x a decimal remainder, it produces an overall delay of (N+1) and takes a weighted average: x of delay N+1 and (1-x) of N. We can model this as a delay of N (which of course doesn't affect the magnitude response) and a two-tap FIR filter providing the remaining portion of the delay. (this graph was made at Fs = 48K; the audio effect is far less at higher sample rates):

As you have observed, the worst case is for a delay of fractional part 0.5 -- where the fractional delay becomes a two-tap, moving average FIR filter. It loses about 12 dB at 20 kHz. It looks terrible by audiophile standards. Yet in a typical application where the delay is part of a variable effect, it may get masked rather nicely. In my case I can't hear the difference (sometimes old age comes in handy).

Best regards,

Bob