Computational Modelling of Yoruba Tonal Grammar for Speech Synthesis Applications

The development of accurate text-to-speech systems for Yoruba requires computational models that capture the complex interaction of lexical tone, downstep, and phrasal intonation, presenting fundamental challenges at the intersection of phonological theory and natural language processing engineering. This dissertation develops original computational models of Yoruba tonal grammar for application in speech synthesis systems, making contributions to both Yoruba formal phonology and African language speech technology. The research combines phonological fieldwork, corpus-based acoustic analysis, and computational modelling in a three-strand methodology. Phonological fieldwork with 40 native speakers in Lagos and Oyo States establishes formal specifications for tone rules, boundary effects, and downdrift parameters. A 20-hour read speech corpus was recorded and annotated for tonal events using a Yoruba-adapted ToBI transcription system. Computational models implementing the formal phonological specifications were developed and evaluated against the corpus annotations and against perceptual acceptability judgements by native speaker evaluators. The theoretical framework integrates Autosegmental-Metrical phonology, target approximation models of tone implementation, and recurrent neural network sequence modelling. Existing computational work on Yoruba speech synthesis has been limited by the absence of a comprehensive formal tonal grammar specification and a large annotated corpus. This dissertation addresses both deficits with original contributions. Findings will demonstrate that a hybrid rule-based and neural modelling approach achieves native speaker acceptability ratings significantly above purely neural baselines for Yoruba tonal synthesis. Recommendations will address open-source release of the Yoruba speech corpus and tonal grammar computational framework. Keywords: computational phonology, Yoruba tonal grammar, speech synthesis, natural language processing, African languages.