[sargun]

Have you looked at applying the techniques used in Google's Wavenet to your corpus? In addition, any interest in releasing your corpus?