We present an approach to mixed initiative dialogue in acoustic user interfaces to databases. First, we discuss how we distinguish between initiative and control in mixed initiative information retrieval dialogue and how the notions of taking, keeping, and relinquishing initiative and control are reflected in our approach. Based on this discussion, we develop a dialogue planning algorithm. This algorithm distinguished between resources and routines and between the type and the content of an utterance; type and content are calculated separately by routines that reason on the resources – a dialogue model, a dialogue history, and an application description. Through this division we achieve a dialogue where the system adapts to the user's attempts at changing the direction of a dialogue. Finally, we argue that automatic segmentation of the dialogue and automatic tracking of initiative and control is inherent to our approach.