Please take a look at function quisk_process_decimate() in quisk.c. Quisk can accept a variety of sample rates even if they are not multiples of 48 ksps. It decimates by an integer to a rate close to 48 ksps. The fractional rate correction happens after demodulation when the audio is mono so there is no problem with images. This applies only to Rx.
I am not sure this helps. If you want I/Q data for a program like WSJT-X it won't work. But many digital programs can work with monophonic data.
If you try to drift the frequency back and forth to average to 48.000 ksps, you will have frequency drifts in WSJT-X.
I must point out that Pico is not a real audio codec.
Jim
N2ADR
