I was under the impression that the audio is stored locally while listening for the trigger word. Once the trigger word is recognized, Alexa begins recording and sends the result to be processed.