Integration of Yoruba language into MaryTTS

Abstract

This paper presents the integration of Yoruba language into MaryTTS and the evaluation of the resulting text-to-speech (TTS) system using the unit-selection method. Yoruba language sounds system contains 54 elements: 12 vowels, 03 tones for each vowel and 18 consonants. The design of our system is based on the extraction of relevant features from the corpus and their use to select the sequences that can give the best synthesized voice output. The evaluation based on the mean opinion score gave us 2.9 out of 5 as score which is equivalent to a good quality. Integration into MaryTTS consisted in adding Yoruba synthesis component to MaryTTS and making Yoruba language available for usage and sharing for research works. Future work will be performed on some challenges such as objective evaluations of output audio stream, expressiveness (which will need prosodic modification) and hidden markov model synthesis for comparison purposes so as to improve the efficiency