Development of a Yoruba Text-to-Speech System Using Festival

Abstract

This paper presents a Text-to-Speech (TTS) synthesis system for Yorúbà language using the open-source Festival TTS engine. Yorúbà being a resource scarce language like most African languages however presents a major challenge to conventional speech synthesis approaches, which typically require large corpora for the training of such system. Speech data were recorded in a quiet environment with a noise cancelling microphone on a typical multimedia computer system using the Speech Filing System software (SFS), analysed and annotated using PRAAT speech processing software. Evaluation of the system was done using the intelligibility and naturalness metrics through mean opinion score. The result shows that the level of intelligibility and naturalness of the system on word-level is 55.56% and 50% respectively, but the system performs poorly for both intelligibility and naturalness test on sentence level. Hence, there is a need for further research to improve the quality of the synthesized speech.

Keywords: Text-to-Speech, Festival, Yorúbà, Syllable