Speech recognition
In developmentKrio ASR
A Whisper model fine-tuned on real Krio speech — voice notes, calls, WhatsApp audio — so everyday conversation transcribes cleanly, not just studio-quality clips.
How we build Krio ASR →Models
Krio language models for speech, translation, and voice.
NovaxAILab builds a small family of models for Sierra Leonean Krio — models that can listen, speak, and translate in a language that most speech systems leave out. Each links to its weights, code, or model page.
Which model should you use?
- Krio ASR
- Transcribing Krio voice notes, phone calls, and field recordings
- Krio TTS
- Reading Krio text aloud in apps, IVR systems, and accessibility tools
- Phos
- Transcribing and translating Krio speech in one step
- Krio Translate
- Written Krio ↔ English translation
Text-to-speech
Krio TTS
A neural text-to-speech model that turns written Krio into natural spoken audio, so a reply is heard the way it would be said in conversation.
Krio TTS on Hugging Face →Speech-to-text with translation
Phos
Transcribes and translates spoken Krio and English in a single workflow, rather than chaining two lossy steps.
Phos model page →Machine translation
In developmentKrio Translate
A dedicated Krio–English translation model built on parallel text and evaluated against Krio references. In active research.
Translation research →Why this matters
The future of AI should not be limited to a handful of globally dominant languages. These models are steps toward speech and language technology that works for education, healthcare, public services, customer support, and accessibility across Africa. Read more in research.