Speech control is an attractive user interface paradigm for many applications. However, there is little support for the systematic construction of the speech interface part of an application. In this paper we introduce a model-based technique for producing a speech interface facility for interactive applications. The technique assumes a conceptual model of the application that is used for inferring the application-specific part of the speech grammar and for the run-time control of the speech interface. The approach has been evaluated by generating an experimental speech interface for Rational Software Modeler and Microsoft Office PowerPoint. The main benefits of the approach are rapid development of speech interfaces, support for customizing applications beyond existing GUI, and support for context-based speech commands.