If you use synthesized speech, your character has the ability to say almost anything, which provides the greatest flexibility. With recorded audio, you can give the character a specific or unique voice. To specify output, provide the spoken text as a parameter of the Speak method.
Because Microsoft Agent's architecture uses Microsoft SAPI for synthesized speech output, you should be able to use any engine that conforms to this specification, and support International Phonetic Alphabet (IPA) output using the Visual method of the ITTSNotifySinkW interface. For further information on the SAPI interfaces, see the Microsoft Speech SDK.