Hohe Speech
Turn spoken Amharic into text. Built for the voices, accents, and expressions of home.
AI models for the way Ethiopia speaks.
Build with Amharic speech recognition today.
Shape what comes next with us.
From the first spoken word to the next big idea.
Explore our available model and the family
we’re developing around it.
Turn spoken Amharic into text. Built for the voices, accents, and expressions of home.
A new voice for your products. Our next chapter: natural speech that sounds closer to home.
Make knowledge travel across languages. Help people connect with content in their own words.
Hohe ASR 1.0 is our first step. Open weights, a public model card, and results you can inspect before you build.
Try speech recognitionRead the model card ↗Reported evaluation: 1,548 held-out clips, 105 speakers. Performance varies by audio type; spontaneous conversation remains challenging. See the model card for methodology and limitations.
Bring your use case. We’ll help you explore
where Ethiopian language AI fits.
Add Amharic transcription to your application. Start with the open model, then talk to us about your integration needs.
Turn spoken content into readable captions. Try our subtitle studio to create and edit Amharic video subtitles.
Explore model evaluation, language data, and adaptation for your organisation’s needs. Let’s define the right pilot together.
Hohe means a letter of the Ge’ez alphabet. A small beginning with the power to carry a whole world of meaning.
We’re building an AI model company around Ethiopian languages, starting with Amharic. Created by the team at Dataset.ET in Addis Ababa, Hohe connects local language expertise with tools people can build on.
Meet Dataset.ET ↗Tell us what you’re building. From a first prototype to an organisation-wide project, we’d like to hear from you.