Researchers have developed new NLP resources for Puno Quechua, an under-resourced language. This includes the largest speech corpus for any Quechua variety, with 66 hours of recordings and 36 hours of transcribed data. The project also established the first ASR benchmark for Puno Quechua, evaluating existing models and fine-tuning Whisper, wav2vec2, and XLS-R. All datasets and fine-tuned models are being released openly to aid language preservation. AI
IMPACT Provides crucial NLP tools and benchmarks for under-resourced languages, potentially enabling new applications and research.
RANK_REASON The cluster contains an academic paper detailing the creation and release of NLP resources for an under-resourced language.
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
