Research on low-resource Tigrinya Automatic Speech Recognition (ASR) systems conducted during my paid Machine Learning Research Internship at Lesan.AI, a German based AI company specifically working on African languages.
This repository contains research materials, experiments, reports, and presentation resources related to the evaluation of low-resource ASR systems for Tigrinya.
- Curated and processed 10,000+ Tigrinya speech utterances
- Conducted ASR benchmarking and evaluation
- Performed linguistic and acoustic error analysis
- Worked with Whisper-based ASR systems
- Contributed to establishing baseline ASR benchmarks for Tigrinya
This work was presented at:
International Conference on Artificial Intelligence for Sustainable Development (AI4SD 2026)
Mekelle University, Ethiopia
April 27–28, 2026
The camera-ready manuscript is under preparation for publication.
docs/ -> Reports and certificates
slides/ -> Presentation slides
notebooks/ -> Research notebooks and experiments
scripts/ -> Supporting Python scripts
Weldesemayat Teklay Gebre
Gebregergs Mekonen