Logo Lanfrica

gabrielmahia/ongea-ai

Domaine:

natural language processing

Type de record:

softwaretools
Créateur:
gab
HĂ´te:
🎙️ Swahili voice AI — speech-to-text, text-to-speech, and voice-first interfaces for low-literacy East African users # 🎙️ Ongea AI — Swahili Voice AI > *Ongea* (Swahili) = speak, talk, converse Voice-first AI interface for Swahili speakers — speech-to-text, text-to-speech, and conversational AI for users who read slowly or not at all. Designed for low-literacy East African users who interact best by voice. ## Why Voice Matters 30% of Kenyan adults read below primary school level. But virtually all own a mobile phone and can speak and listen. Voice AI removes the literacy barrier — users speak their question in Swahili and hear the answer. No typing, no reading, no language barrier. ## Architecture ``` User speaks (Swahili) → Whisper (speech-to-text) → Gemini (reasoning in Swahili) → TTS (text-to-speech, Swahili) → User hears answer ``` ## Use Cases - 🌾 Farmer asks: "Je, leo ni wakati mzuri wa kupanda mahindi?" - 💊 Patient asks: "Dawa hii ninaitumia vipi?" - 💰 Business owner: "Ninaweza kupata mkopo mdogo vipi?" - 📋 Voter: "Namshinda nani mgombea katika kura yangu?" ## Quickstart ```bash pip install -r requirements.txt streamlit run app.py ``` ## Research Basis OpenAI Whisper (Radford et al. 2022) has documented support for Swahili. The voice-first interface pattern is validated by USSD and SMS mHealth research in East Africa (PMC10562967). --- *gabrielmahia.ai | Part of the Swahili AI Infrastructure Stack*