OkwuGbé: End-to-End Speech Recognition for Fon and Igbo
Abstract
Language is a fundamental component of human communication. African low-resourced languages have recently been a major subject of research in machine translation, and other text-based areas of NLP. However, there is still very little comparable research in speech recognition for African languages. OkwuGbé is a step towards building speech recognition systems for African low-resourced languages. Using Fon and Igbo as our case study, we build two end-to-end deep neural network-based speech recognition models. We present a state-of-the-art automatic speech recognition (ASR) model for Fon, and a benchmark ASR model result for Igbo. Our findings serve both as a guide for future NLP research for Fon and Igbo in particular, and the creation of speech recognition models for other African low-resourced languages in general. The Fon and Igbo models source code have been made publicly available. Moreover, Okwugbe, a python library has been created to make easier the process of ASR model building and training.- Anthology ID:
- 2021.winlp-1.1
- Volume:
- Proceedings of the Fifth Workshop on Widening Natural Language Processing
- Month:
- November
- Year:
- 2021
- Address:
- Punta Cana, Dominican Republic
- Venue:
- WiNLP
- SIG:
- Publisher:
- Association for Computational Linguistics
- Note:
- Pages:
- 1–4
- Language:
- URL:
- https://aclanthology.org/2021.winlp-1.1
- DOI:
- Cite (ACL):
- Bonaventure F. P. Dossou and Chris Chinenye Emezue. 2021. OkwuGbé: End-to-End Speech Recognition for Fon and Igbo. In Proceedings of the Fifth Workshop on Widening Natural Language Processing, pages 1–4, Punta Cana, Dominican Republic. Association for Computational Linguistics.
- Cite (Informal):
- OkwuGbé: End-to-End Speech Recognition for Fon and Igbo (Dossou & Emezue, WiNLP 2021)