Seems like this attack could be trivially prevented by disabling speech recognition while speech synthesis is running The paper mentions that they save the syn

Seems like this attack could be trivially prevented by disabling speech recognition while speech synthesis is running The paper mentions that they save the synthetic voice to .wav files first, and only later play them back, so this would not help.