This is not a supported feature in AVS.
Depending on what you are trying to accomplish here you could potentially use the Alexa Skills Kit and make your own customer skills which allows you to specify an utterance to be returned to the client. Then you would send a speech request to AVS along the lines of "Ask my custom skill to blah" and then you would receive a SpeechSynthesizer directive with audio corresponding to the utterance that you have coded in your custom skill.
There is more information on the skills kit here:
iOS: Downchannel confusion 1 Answer
Invalid API Key for Android Sample app 5 Answers