Newest papers

AfriBED: a Speech Recognition Bias Evaluation Dataset for Low Resource African Languages

Claytone Sikasote, Hussein Suleman, Jan Buys · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.26821 · cited by 0 works (OpenAlex) · open-access PDF

Indonesian

Meskipun kemajuan dalam sistem pengenalan bicara otomatis (ASR), sistem saat ini menunjukkan bias atribut pembicara, yaitu, kinerja pengenalan bicara mereka bias terhadap pembicara tertentu berdasarkan atribut demografis mereka seperti gender, aksen, dan jenis pembicara. Namun, dengan tepat menilai bias atribut pembicara dalam sistem ASR tertentu memerlukan ketersediaan dataset evaluasi bias dengan pembicara yang sebanding di seluruh subgrup atribut pembicara. Dalam makalah ini, kami memperkenalkan African Speech Bias Evaluation Dataset (AfriBED), sebuah dataset penilaian bias pengenalan bicara novel untuk dua bahasa Afrika sumber daya rendah, Bemba dan Nyanja, untuk mendeteksi dan memperkirakan bias jenis pembicara dan jenis pembicara untuk sistem ASR. Dataset ini terdiri dari 9 jam bicara dengan 3928 utter

Machine translation (Meta M2M100); the English original follows.

Abstract

Despite advancements in automatic speech recognition systems (ASR), current systems exhibit speaker-attribute bias, i.e., their speech recognition performance is biased against certain speakers based on their demographic attributes such as gender, accents, and speaker type. However, accurately assessing speaker attribute bias in a given ASR system requires the availability of a bias evaluation dataset with comparable speech across subgroups of a speaker attribute. In this paper, we introduce African speech Bias Evaluation Dataset (AfriBED), a novel speech recognition bias evaluation dataset for two low resource African languages, Bemba and Nyanja, to detect and estimate gender and speaker-type bias for ASR systems. The dataset comprises 9 hours of speech with 3928 utterances recorded by 80 speakers for Bemba, while the Nyanja set includes 6260 utterances with approximately 18 hours of speech recorded by 127 speakers. The results of our baseline experiments demonstrate that the datasets are useful for examining gender and speaker-type attribute bias in the target languages.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Computer science, Speech recognition, Set (abstract data type), Speaker recognition, Voice activity detection, Speech processing

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange