Optimisation of multiple feature stream weights for distributed speech processing in mobile environments