Building Community-Centred NLP Resources for Puno Quechua

2026-05-27

AI Summary

This study addresses the critical scarcity of digital speech resources and automatic speech recognition (ASR) tools for Puruhá Kichwa, a low-resource Indigenous language, which has hindered its preservation and revitalization. Through a community-engaged design approach, the authors present the first large-scale speech corpus for the language—comprising 66 hours of recorded speech, with 36 hours manually transcribed and validated—and establish the first systematic ASR benchmark. Fine-tuning state-of-the-art models (Whisper-base, wav2vec2-base, and XLS-R-300M) with continued pretraining yields the current best performance. All data and models are fully open-sourced, promoting a community-led, community-serving paradigm for linguistic resource development.

Abstract

The preservation of under-resourced languages requires digital tools and resources shaped by and for their speakers. We present the first dedicated ASR resources for Puno Quechua (ISO 639-3: qxp): (1) the largest speech corpus for any single Quechua variety, consisting in 66 hours of recordings for scripted and spontaneous speech (including 36 hours of manually transcribed and validated data), collected via a participatory design campaign; (2) the first systematic ASR benchmark for Puno Quechua, evaluating state-of-the-art models and fine-tuning Whisper-base, wav2vec2-base, and XLS-R-300M, with and without continued pre-training (CPT); (3) an open release of all datasets and fine-tuned models.

Problem

• low-resource languages
• speech corpus
• automatic speech recognition
• Quechua
• language preservation

Innovation

• Puno Quechua
• automatic speech recognition
• low-resource languages
• participatory design
• speech corpus