Improved Syllable-Based Text to Speech Synthesis for Tone Language Systems

In this contribution, we document the series of progress towards attaining a generic and replicable system that is applicable not only to Nigerian languages but also other African languages. The current system implements a state-of-the-art approach called the Hidden Markov Model (HMM) approach and aims at a hybridised version which front end components would serve other NLP tasks, as well as future research and developments. We continue to tackle the language specific problems and the 'unity of purpose' phenomenon for tone language systems and improve on the speech quality as an extension of our LTC' 2011 paper. Specifically, we address issues bordering on tone modelling using syllables as basic synthesis units, with an 'eye ball' assessment of the synthesised speech quality. The results of this research offer hope for further improvements, and we envisage an unsupervised system to minimise the labour intensive aspects of the current design. Also, with the active collaboration network established in the course of this research, we are certain that a more robust system that would serve a wide variety of applications will evolve.