01
Data collection
Voice and text data are gathered from native speakers, public sources, and curated multilingual content to build a broad and representative foundation.
Research · Publications
Publications on speech, translation, and African language AI.
Our research write-ups on automatic speech recognition, neural text-to-speech, and machine translation for low-resource African languages — along with the data pipeline that makes them possible.
Data pipeline
From raw speech and text to usable training data.
Every model begins with a pipeline that is designed to be repeatable, transparent, and practical for low-resource languages. The process spans collection, cleaning, annotation, evaluation, and release.
02
Curation and cleaning
Audio is normalized, segmented, and quality-checked while text is standardized and aligned to reduce noise and improve learning quality.
03
Annotation and alignment
Transcripts, translations, and metadata are verified so the data is reliable enough for training and evaluation.
04
Training and evaluation
Models are fine-tuned, validated, and benchmarked with both automated metrics and human review to measure real-world usefulness.
05
Release and iteration
The best-performing models and datasets are packaged for reuse, while feedback from deployment helps shape the next iteration.
Publications
Research papers and technical reports.
These publications capture the research direction behind the data pipeline, model development, and language infrastructure work.
Research on Automatic Speech Recognition for Low-Resource African Languages
A detailed study of the development of a Krio speech-to-text system, including data collection, preprocessing, model training, and evaluation.
Read more →Research on Text-to-Speech for Low-Resource African Languages
An exploration of neural TTS methods for Krio, focused on naturalness, data preparation, and extension to other African languages.
Read more →Research on Neural Machine Translation for Low-Resource African Languages
A multilingual translation study beginning with Krio and outlining a broader path for African language translation systems.
Read more →