A voice activity detection (VAD) dependent retransmission scheme is described that mitigates the effect of packet loss on an audio signal transmitted between terminals in a wireless communication system in a manner that is generally more robust than conventional state-of-the art packet loss concealment algorithms but that consumes less terminal power as compared to conventional retransmission schemes. In one implementation, this is achieved by allowing retransmissions to be requested by a terminal only when a packet received by the terminal is deemed bad and when a portion of an audio signal currently being received by the terminal is deemed to comprise active speech. In other implementations, the processing of retransmission requests received by a terminal is inhibited or turned off entirely during periods when a portion of an audio signal currently being transmitted by the terminal is deemed not to comprise active speech.