I believe it is practical to support 64kbps, and some simple compression can be done (if two 10 bit samples are the same, which would happen somewhat frequently, don't transmit it again. Can be done with a little overhead by simply referencing which samples to duplicate at the start of each chunk).
In any case, my biggest question is if I can sample the audio from a mic using ADC, and on the other device send the output using the DAC of the ATMega microprocessor to a speaker and it would play.