I have a problem with the final release. The answers given by the assistant are not the robot voice, but its own audio file (in SSML).
An example of the SSML code: