Neural Text-to-Speech Systems
My post-graduate transfer thesis; unfortunately I never got the chance to follow up on some of these ideas. An abridged abstract:
"We believe that there are 3 issues affecting the progress of Neural Text-to-Speech (TTS) synthesis: first, an under-utilization of existing phonetic and linguistic knowledge. Second, a systemic lack of rigour in the reporting and analysis of results. Third, an unnecessary and wasteful increase in model sizes. This report gives a review of relevant literature, details of research carried out during the 18-month period, and points to future lines of investigation."