Tag
This research provides critical insights into optimizing AI architectures for Automatic Speech Recognition (ASR) in low-resource African languages. It offers practical guidance…
This research critically evaluates the safety of large language models in Yoruba and Nigerian Pidgin, revealing that English-centric safety measures fail to protect users in these…
This research is a foundational study on safety alignment for large language models specifically for Hausa, a major West African language. It introduces a new Hausa safety dataset…
This research introduces SunflowerASR, a new speech recognition model optimized for low-resource African languages using Group Relative Policy Optimisation. It demonstrates…
AfriBED is a new speech recognition bias evaluation dataset created for two low-resource African languages, Bemba and Nyanja. This dataset enables the detection and estimation of…
A new dataset, AfriBED, has been created to evaluate and reduce speaker-attribute bias in Automatic Speech Recognition (ASR) systems for low-resource African languages,…
This research introduces TTSYoruba, a specialized text-to-speech system for the Yoruba language, integrated into an online dictionary. It significantly improves the digital…
Pearl-Chat is a new 8.5-million parameter language model specifically engineered for Luganda, a low-resource Bantu language spoken by over 10 million people in East Africa. This…
TTSYoruba is a new speech synthesizer specifically designed for the Yoruba language, a significant step in developing AI tools for low-resource African languages. It is deployed…
A new 174-hour Ewe ASR benchmark, 'Nufo', has been released, providing a crucial resource for developing speech recognition technology for this major West African language. This…
A new rule-based AI speech synthesizer, TTSYoruba, has been developed specifically for the Yoruba language, addressing its complex tonal and phonetic features. This system is…
This paper proposes "Model as a Library" (MaaL), an AI architecture specifically designed to address the challenges of language diversity and low digital literacy in African…
This research introduces MGhana-ST, a new speech translation dataset for four low-resource Ghanaian languages, significantly enhancing AI capabilities for African languages. The…
This research introduces RunyaNER, the first publicly available Named Entity Recognition (NER) benchmark for Runyankore, an East African language. This dataset and the…
This research directly addresses the challenge of deploying reliable AI text classifiers in low-resource African languages. It demonstrates that language-specific confidence…
This research directly addresses the performance gap of large language models for low-resource languages by focusing on ten African languages. It proposes geometric regularization…
A new AI model, VEXMLM, significantly improves language processing for Ge'ez-script African languages like Amharic and Tigrinya. These advancements also transfer to 17 other…
This research directly addresses the challenges of developing automatic speech recognition for low-resource African languages. The study utilized two Africa-centric datasets to…
A new rule-based AI speech synthesizer, TTSYoruba, has been developed for the Yoruba language, a significant step for African language technology. This system is deployed online…
Researchers have developed TTSYoruba, an advanced text-to-speech system specifically for the Yoruba language, a major language spoken across West Africa. This innovation addresses…
This research significantly advances AI's capability to adapt multilingual models to low-resource languages by evaluating its novel DeltaMerge-LowRes method on four African…
This research presents the first end-to-end Text-to-Speech (TTS) study for Efik, a tonal language spoken in Southeastern Nigeria, addressing its underrepresentation in digital…
A new study exploring Automatic Speech Recognition for low-resource African languages found that linguistic relatedness alone is not a reliable predictor for cross-lingual…
A new AI technique, Crosslingual On-Policy Self-Distillation (COPSD), has been developed and proven to significantly enhance mathematical reasoning capabilities of large language…
