FLEURS: Few-shot Learning Evaluation of Universal Representations of Speech
Speech version of FLoRes with ~2,009 n-way parallel sentences per language read by native speakers, for ASR, ST and language ID evaluation.
Search titles, authors, subjects and dates from across the DARE African Knowledge Repository.
73 resources found in the repository.
Speech version of FLoRes with ~2,009 n-way parallel sentences per language read by native speakers, for ASR, ST and language ID evaluation.
Largest MT benchmark for African languages in the news domain, with train/dev/test English-or-French to African-language translation pairs.
Practical notebooks covering Bayesian inference, deep generative models, graph neural networks, JAX, reinforcement learning and transformers.
Bilingual practical notebooks from Indaba 2024 covering JAX, probabilistic programming, GNNs, responsible AI, LLMs, diffusion models, federated learning, low-resource NLP and recommender systems.
Shona-adapted mT5 small model for conversational AI, machine translation, summarisation, question answering and NER.
Practical notebooks from the 2025 Deep Learning Indaba, the annual gathering of the African machine-learning community.
Parallel English-Shona sentences derived from OPUS MT560 for machine translation, cross-lingual NLP and language-model fine-tuning.
Audio-transcription parallel data for chiShona tagged for translation, conversational, token-classification and text-generation tasks.
Human-annotated multilingual instruction-tuning dataset curated via an open-science annotation programme with substantial African-language participation.
Large multilingual instruction dataset collection assembled for the Aya project, aggregating open datasets across 101 languages.
Distilled 600M-parameter multilingual translation model covering 200 languages, including a large number of African languages.
Curated list of Amharic stopwords generated and validated by the LT Group at the University of Hamburg.