Introducing Phos: speech-to-text with translation for African languages.
Phos is designed to listen, transcribe, and translate in ways that are useful for real communities — from customer support to education and public services. It becomes available on August 1, 2026.
One workflow, not two
Most speech pipelines treat transcription and translation as separate steps, which means audio in a low-resource language has to survive two lossy hand-offs before anyone can act on it. Phos is a speech-to-text model with translation built in, so spoken language is transcribed and meaningfully converted into another language in a single workflow.
What it can do
- Speech-to-text transcription for low-resource African languages
- Translation across Krio and English
- Real-time conversational workflows for accessibility and service delivery
- Multilingual understanding for education, healthcare, and public services
Why Krio first
Krio is spoken across Sierra Leone as an everyday lingua franca, yet it is almost entirely absent from commercial speech and translation systems. Starting there means the model is measured against how people actually speak — voice notes, phone calls, code-switching and informal speech — rather than against studio recordings in high-resource languages.
Availability
Phos is available from August 1, 2026. You can read the full model details on the Phos model page, see how it fits into our wider research on African language AI, or review API pricing.