Evaluating generalised additive mixed modelling strategies for dynamic speech analysis