There is a helpful
app-note from Atmel that describes how to handle audio recording and playback with an AVR microcontroller, using the built in ADC for acquisition and PWM for playback. In your case, just imagine a radio link to another AVR in place of the SPI data flash shown in the app-note.
I used this for adding a voice mode to a GPS weather balloon tracking system. This enabled it to read out a radio callsign and GPS coordinates in English, from voice samples stored into the AVRs program memory as 8-bit PCM at a 4kHz sample rate. The sample rate was originally 8kHz, which sounded much clearer, but exceeded the limited program memory of the AVR and had to be reduced. At 4kHz the voice is still intelligible (at least to me), though my friends joke that it sounds like Charlie Brown's teacher.