In this paper, we extend the persona-based sequence-to-sequence (Seq2Seq)\nneural network conversation model to a multi-turn dialogue scenario by\nmodifying the state-of-the-art hredGAN architecture to simultaneously capture\nutterance attributes such as speaker identity, dialogue topic, speaker\nsentiments and so on. The proposed system, phredGAN has a persona-based HRED\ngenerator (PHRED) and a conditional discriminator. We also explore two\napproaches to accomplish the conditional discriminator: (1) phredGAN_a, a\nsystem that passes the attribute representation as an additional input into a\ntraditional adversarial discriminator, and (2) phredGAN_d, a dual discriminator\nsystem which in addition to the adversarial discriminator, collaboratively\npredicts the attribute(s) that generated the input utterance. To demonstrate\nthe superior performance of phredGAN over the persona Seq2Seq model, we\nexperiment with two conversational datasets, the Ubuntu Dialogue Corpus (UDC)\nand TV series transcripts from the Big Bang Theory and Friends. Performance\ncomparison is made with respect to a variety of quantitative measures as well\nas crowd-sourced human evaluation. We also explore the trade-offs from using\neither variant of phredGAN on datasets with many but weak attribute modalities\n(such as with Big Bang Theory and Friends) and ones with few but strong\nattribute modalities (customer-agent interactions in Ubuntu dataset).\n