Voice and multimedia applications in wireless sensor nodes are a relatively new topic with a wide array of applications such as emergency speech communications and video monitoring. Although this topic has been throughly investigated with respect to traditional networks, multimedia in WSNs presents challenges of low processing capabilities and energy constraints, all of which must be taken into consideration. The wireless nature of the communication also causes a severe degradation of the intelligibility of the speech data due to packet loss. In this presentation we will present a perceptual marking technique which can be used to determine the most important speech packets, allowing us to achieve higher speech quality with lower energy consumption compared to other techniques by selectively retransmitting or cooperating, thus to enable important packets to reach their destination.