Resources
A curated library of Sunbird AI’s published research, policy briefs, annual reports, datasets, and technical documentation.
Sunflower: A New Approach to Expanding Coverage of African Languages in Large Language Models
Benjamin Akera et al., Sunbird AI, 2025. Describes Sunflower 14B and 32B — open-source LLMs with state-of-the-art performance in 31 Ugandan languages.
SALT: A Multi-way Parallel Corpus for Ugandan Languages
Technical report on the SALT dataset — 25,000 sentences across 6 languages, used to train Sunbird’s translation and speech systems.
Feature Learning for Efficient ASR-free Keyword Spotting in Low-Resource Languages
Sunbird research on speech recognition for Ugandan radio monitoring.
Sunbird AI Annual Report 2022
A review of Sunbird’s projects, partnerships, and milestones in 2022.
Sunbird AI Annual Report 2020
Sunbird’s founding year report — covering the first SALT corpus and the establishment of the organisation.
Feature Learning for Efficient ASR-free Keyword Spotting in Low-Resource Languages
Sunbird research on speech recognition for Ugandan radio monitoring.
From Complaint to Evidence
Policy Brief a data foundation for noise goverance in Uganda
GRAIN: Gender and Responsible AI Network
Policy framework and research agenda of the GRAIN consortium for gender-responsive AI in sub-Saharan Africa.
SALT Dataset — Multi-way Parallel Corpus of Ugandan Languages
25,000+ sentence parallel text corpus in 6 languages under CC-BY license. Includes studio speech recordings.
Urban Noise Dataset — Kampala Environmental Sounds (19 Categories)
Open environmental noise dataset from Sunbird’s sensor network across Kampala.
Sunbird AI Model Hub on HuggingFace
All open-source models: Sunflower 14B and 32B, translation models (NLLB-SALT), ASR and TTS systems. Free to download, adapt, and deploy.
Sunbird AI GitHub — Open Source Repositories
All Sunbird code: SALT tools, API source, noise monitoring systems, ASR portal, and more.