The great advantage of this method is that the behavior of the speaker is very close to that of future users of the intended application. Furthermore, different design aspects of the application can be `tested' beforehand and data may be analyzed to model user reactions in the application more successfully.
Depending on the effort that is put into the recording setup the acoustical environment can be matched very closely to that of a real-world situation. Therefore the data collected in WOZ technique is usually the best you can get for a complex application.
On the other hand WOZ recordings require a much higher effort in costs and man power: The setting must be so convincing that naive users do not suspect they are being tricked, the recording itself requires as a minimum two persons (one supervisor and one wizard), and finally because the speaker is `steered' by the simulated system, WOZ requires a lot more training of the persons who do the recordings.
WOZ recordings may be designed in many different settings depending on your needs; therefore it is difficult to give detailed instructions on how to specify them. Also, many problems cannot be foreseen because WOZ recordings are definitely not standardized recordings. The best we can advise you here is that you stay as unspecific about the WOZ technique in the specs as possible. Try to concentrate on the overall intention - for instance that the users must not be aware of the simulation, that the setup matches the real situation as well as possible, and so on - but do not give any hard facts. On the other hand try to get as much information into the specifications as possible about the intended application. This is the basis you have to work on; if you do not exactly know how the `virtual machine' has to work, you're lost. This is very important if you produce the speech corpus as a contractor.
See the section (p. ) for some more technical hints for the WOZ technique. In the section (p. ) you will find as an example a rather complicated WOZ recording setup used in the German SmartKom project.