the doc sucks. i'm playing around with this API as well. according to sample code, all parameters are in bytes , which means:

samples = bytes/ [bytes per sample] / channels

I'm doing FFT using adc & i2s, with sample rate 8K and 16K, the result is not correct. by setting sample rate to 44100, everything works fine.

for 16bits data, words swap is mandatory as well.