Providing real-time speech support over multi-hop ad hoc wireless networks is a challenging task. In order to make a voice application to be feasible over ad hoc wireless networks, the perceived voice quality must be improved while reducing the packet overhead. In this work, we propose various mechanisms to achieve this objective. Using these mechanisms, we propose an efficient packetization scheme, in which, the important sub-stream of the voice stream is protected adaptively with forward error correction (FEC) depending upon the network state and is transmitted over two disjoint paths. The less-important sub-stream of the voice stream is encoded into two descriptions, which are then transmitted over two maximally node disjoint paths. We derived an expression for estimating residual packet loss rate, RPLR for the given FEC-offset, r (the distance between original voice frame and piggybacked redundant voice frame) and packet loss in the network. Our scheme adapts the FEC-offset value (it chooses the FEC-offset that minimizes RPLR as much as possible) based on loss rate feedback obtained from the destination. As observed from simulations, the proposed scheme achieves significant gains in terms of reduced frame loss rate, reduced control overhead, and minimum end-to-end delay and almost double the perceived voice quality compared to the existing approaches