Abstract: Method and apparatus for modifying a speech input to generate a desired speech output. A voice recognition unit analyzes the speech input and generates a textual output. The output is used by a speech synthesis unit to generate an output speech signal using speech characteristics and parameters stored in a database. The database may be preconfigured to store speech characteristics of a variety of types. A training unit allows the user to enter speech samples which are used to generate inputs to the database. The training unit acquires the speech samples and generates speech units, such as diphones, which may be modified to achieve a desired result. In one embodiment, the modified speech output is applied to an STU program, and feedback is provided indicating the error rate.