New AI System Delivers High-Quality Yoruba Speech Synthesis

Researchers have developed TTSYoruba, a rule-based concatenative diphone speech synthesizer specifically for the Yoruba language. This system, integrated into the YorubaName.com online dictionary, converts tone-marked Yoruba text into audio using a meticulously crafted phonological rule system and a library of 651 recorded diphone units. The project addresses the unique linguistic complexities of Yoruba, including its five tonal variants and specific challenges like nasal disambiguation.

The paper details the architectural design of TTSYoruba, focusing on its phonological rules, tone selection logic, and the handling of nasal sounds. A significant orthographic contribution is the standardization of caron and circumflex symbols for marking contour tones in single vowels, which are integrated into the system's text normalization and the WriteYoruba keyboard tool. This innovation aims to improve the accuracy and usability of written Yoruba for speech synthesis.

The system's effectiveness was validated through a listener study involving 50 participants, with detailed Mean Opinion Scores (MOS) provided in the research. This work represents a crucial advancement in text-to-speech technology for low-resource African languages, offering a robust solution for a language spoken by millions across West Africa and the diaspora. It highlights the potential of AI to preserve and promote linguistic diversity.