Speech-synthesis technology designed for practical utility
Time: Tue 2021-11-30 15.15
Location: Fantum and Zoom
Participating: Shivam Mehta
In this seminar, I will discuss the progress achieved during the first
year of my PhD studies. The main contribution is a novel probabilistic
method of synthesising speech by combining the monotonic alignment
capabilities of an HMM with the better acoustic modelling of the recent
neural sequence-to-sequence text-to-speech (TTS) systems. We call the
resulting method "neural HMM TTS". In the presentation, I will briefly
reflect upon the motivation of the research and the results of our
proposed method along with demonstrations of synthesised speech. I will
also address some implementational challenges and plans for the second
year of my PhD.