Speaker | Building Community-Centred AI

Quechua is spoken by millions across the Andes but is almost absent from voice-driven AI. There is very little annotated speech data and speech models perform poorly. Existing tools push a largely oral community into text-heavy interfaces that don’t match how people actually communicate. QuechuaBase.org (QB) is an open infrastructure, curated by the community, for oral-first language technology: speech recognition (ASR), text-to-speech (TTS), speech emotion recognition (SER) and open educational resources (OER). With local partners in Puno we have collected 65h+ of speech and we aim for 100h+ by the end of 2026. QB will support voice assistants, reading apps for all Quechua varieties, and community-owned archives. It also offers a blueprint other minority languages can follow.