Optimizing speech naturalness in voice user interface design: A weakly-supervised approach

Speech-enabled applications are gradually finding their way into the lives of today's society. One major concern that is likely to increase the acceptability and popularity of these products is how natural the products sound. This paper therefore concentrates on improving speech naturalness in voice user interface (VUI) designs. The reason for this is to enhance communication, as users will be interacting with a more human (natural) sounding system. Full implementation of a tone language synthesizer, using the Hidden Markov Model approach is presented. Results obtained from this implementation sounds promising with improved intelligibility and well-spoken utterances. The system is currently undergoing rigorous refinements to ensure adaptability and robustness of the design. We are optimistic that this paper shall expand our science and technology horizons and prepare us, the developing world for an in-depth research into the field of speech technologies that will certainly re-direct our livelihood.