Hi, I’m Yacouba Diarra (diarray on the internet) and I care about stuffs. And, by the way, this website was created with markdown, what a great “framework” to create a motherfucking website!

Right now? I’m doing research on speech recognition at RobotsMali AI4D Lab. If you want to know more about me check my ChatGPT biography @ whoami, I’ll try to update it maybe once every leap year. Here are some of the stuffs I have worked on and places you can find them (I’ll update this more often):

Projects

2024 - Present

Research Engineer @ RobotsMali AI4D Lab.

Datasets & Models: Released African Next Voices (612h) and Kunkado (160h) open Bambara ASR datasets and associated models
An be Kalan: Initiated the RobotsMali literacy apps development (GitHub / Store) + Quantization & Edge deployment of our ASR models to enable offline usage in Mali.
Fintech: Lately, my focus shifted to exploring and adapting foundational ML techniques and our speech foundations to drive financial inclusion in Mali.

2022 - Present

Personal Projects

Space: I am currently exploring Machine Learning for Observational Astronomical and space science. Lately, I've been studying NASA and IBM Heliophysics foundation Model
Robotics: ROS and Computer Vision; bronze medalist at the 2023 Panafrican Robotics Competition.
Credit Scoring: I developed fundamental skills during my undergrad project, using DFS to create features from a large relational database and compare ML algorithms for credit scoring.
Learning: At some point I was obsessed about implementing DL techniques from scratch; keep in mind this was started in the pre-ChatGPT era so it was all struggle and learnings. Special mention to Michael Nielsen for his book.

Publications

For a full list of my publications, check my Semantic Scholar or Google Scholar profiles.

Listen, Attend, Understand: a Regularization Technique for Stable E2E Speech Translation Training on High Variance labels (2026). Yacouba Diarra and Michael Leventhal. arXiv
Kunnafonidilaw ka Cadeau (Kunkado): an ASR dataset of present-day Bambara (2025). Yacouba Diarra, P. A. Kamaté, N. S. Coulibaly, and M. Leventhal. AfricaNLP @ EACL 2026 | arXiv | HuggingFace
Dealing with the Hard Facts of Low-Resource African NLP (2025). Yacouba Diarra et al. AfricaNLP @ EACL 2026 | arXiv | HuggingFace
Where Are We At with Automatic Speech Recognition for the Bambara Language? (2026). Yacouba Diarra et al. AfricaNLP @ EACL 2026 | arXiv
Cost Analysis of Human-corrected Transcription for Predominately Oral Languages (2025). Yacouba Diarra, N. S. Coulibaly, and M. Leventhal. arXiv

Blog

Because I may occasionally write stuffs that do not quite fit in an academic paper, I started a blog. This blog really doesn’t have a specific topic, the books I read are way too random for that, and I think it will often be about a book. So let’s say it’s a reading blog.