Ghana NLP / Khaya AI - UNICEF WCARO NLP

Organization Information

TypeResearch Community

Founded2020

Sizelarge

Maturityestablished

Opennesspartial

LocationGhana (distributed team)

FundingVolunteer-driven nonprofit. Grant support from Hugging Face, Google, Microsoft. Commercial arms: Algorine Inc. (research lab) and Khaya AI (enterprise API services).

Websitehttps://ghananlp.org/

GitHubhttps://github.com/GhanaNLP

HuggingFaceGhana-NLP

Contactnatural.language.processing.gh@gmail.com

Coverage

CountriesGhana, Nigeria, Kenya

Languages

Twi Ewe Ga DagbanigurkusYorubakik

Description

Ghana NLP is an open-source initiative focused on Natural Language Processing for Ghanaian and African languages. They developed Khaya, the first AI-powered translation app for Ghanaian languages, which has expanded to cover 17 languages including Kenyan and Nigerian languages. Their translation models outperform Google Translate for Ghanaian languages. The name "Khaya" comes from the African Mahogany tree and means "home" in several Southern African languages. The team now also operates Khaya AI (khaya.ai) as a commercial startup offering enterprise API access for African language translation, ASR, and TTS.

🦄 UNICEF Relevance

(Auto generated) Directly relevant for UNICEF Ghana's AGOO platform and Asempa AI companion, which users are requesting in local languages. Khaya already supports all major Ghanaian languages (Twi, Dagbani, Ewe, Ga) that AGOO uses. Their API could be integrated into existing UNICEF platforms. Translation + ASR + TTS stack enables voice-first interfaces for low-literacy users. Volunteer-driven model means potential for co-development partnerships. Also covers Yoruba (Nigeria) which could support other country offices.

Key People

Dr. Paul Azunre - Co-founder, Director of Research (PhD MIT, Author "Transfer Learning for NLP")
Founded Algorine Inc. research lab. Led development of world's first MT and speech systems for Twi, Ewe, Ga, Dagbani, Gurene. Principal investigator on DARPA programs.

Projects

Khaya Translator (web + mobile app): AI-powered translation for 10+ African languages. Web, Android, iOS apps available. Outperforms Google Translate for Ghanaian languages. ASR capability.
Twi Ewe Ga DagbanigurkusYorubakik
Khaya API: Developer API for integrating African language translation, ASR and TTS into apps.
Twi Ewe Ga DagbanigurkusYorubakik
Khaya ASR: Automatic speech recognition for Twi, Ewe, Ga, Dagbani, Yoruba. Speak to phone and get text transcription.
Twi Ewe Ga Dagbani Yoruba
Khaya TTS: Text-to-speech for 36 African languages
Twi Ewe Ga Dagbani Yoruba Hausa Igboswaamh
ABENA: BERT-based NLP model for Twi language. (2017)
Twi
GhanaNLP Parallel Corpora: 41,513 human-translated parallel sentence pairs (English↔local language) across 5 Ghanaian languages: Twi (14,875), Ga (11,652), Fante (5,001), Ewe (4,985), Kusaal (5,000). Multi-domain (health, education, conversation, religion), dialect-aware. CC BY-NC-SA 4.0. Google-funded. Used to train Khaya AI translation engine.
Twi Ewe Gakus
Naafi TTS (Ewe): Text-to-speech demo for Ewe language.
Ewe

View all projects →

Publications

GhanaNLP Parallel Corpora: Comprehensive Multilingual Resources for Low-Resource Ghanaian Languages (2026)

View all publications →

Partnerships

Google
Microsoft
Harvard University
Algorine Inc.

Notes

(Auto generated) Strong technical team led by MIT PhD. Production-ready apps on App Store and Play Store. API available for integration. Outperforms Google Translate for covered languages. 105 contributors shows strong community. Expanding beyond Ghana to Kenya and Nigeria demonstrates scalability. Could be key partner for UNICEF Ghana's language tech needs.