RIYE Audio Dataset: A Multidialectal Speech Corpus for Low-Resource Language Processing

Abstract

This dataset consists of a curated collection of high-fidelity, field-recorded audio samples developed under the Digiculture RIYE Project's ethnographic survey framework. Created to bridge the digital divide for under-resourced languages, this corpus captures diverse regional speech patterns, unique tonal variations, and distinct phonetic markers essential for localised speech research. The dataset is designed to support a wide array of machine learning, signal processing, and computational linguistics tasks. Because the audio is formatted for clean feature engineering, it serves as an ideal baseline for researchers developing neural networks, automatic speech recognition (ASR) engines, and lightweight audio classification systems optimised for edge deployment.