Simulate conversation with using two different voices?

Unfortunately, no.
Amazon Polly is only configured to do one voice at a time. The developers decided not to support the SSML <voice> tag which would have made it easy to change voices for one encoding. Microsoft and Google support this so perhaps one day Amazon Polly will, too?

For now, you can manually combine two .mp3 files together relatively easily through a Command line. If you do a Bing/Google search you'll find the commands to do this.