Poster | Building Community-Centred AI: Quechua Languages

Preserving endangered languages requires more than translation; it demands digital tools that respect oral traditions. Quechua, spoken by millions across the Andes, remains largely absent from voice-driven AI, forcing speakers into text-heavy interfaces that ignore community practices. Here we present a community-centred AI framework built on Quechua Base (QB), an open infrastructure for modelling, curating and deploying oral-first language technologies. By addressing ASR, TTS, and SER, our approach shifts from text-centric to oral-centric AI, providing a scalable blueprint for technology that genuinely serves minority language communities. For Puno Quechua (qxp) we collected 65h+ of speech and 5k+ sentences, aiming for 100h+ and 10k+ by the end of 2026, and propose XLS-R-300M with continued pre-training (CPT) for ASR.