Save memory by setting up voice data on demand, instead of allocating buffers for each player object