AfriHuBERT: A self-supervised speech representation model for African languages

TL;DR

AfriHuBERT is a self-supervised speech model trained on over 1,200 African languages, significantly improving speech recognition and language identification performance for African languages, and demonstrating strong generalization and low-resource capabilities.

Contribution

This work extends mHuBERT-147 to include over 1,200 African languages by large-scale pretraining, advancing speech representation for underrepresented languages.

Findings

3.6% F1 score improvement in SLID

2.1% WER reduction in ASR

Competitive with larger SSL models

Abstract

In this work, we present AfriHuBERT, an extension of mHuBERT-147, a compact self-supervised learning (SSL) model pretrained on 147 languages. While mHuBERT-147 covered 16 African languages, we expand this to 1,226 through continued pretraining on 10K+ hours of speech data from diverse sources, benefiting an African population of over 600M. We evaluate AfriHuBERT on two key speech tasks, Spoken Language Identification (SLID) and Automatic Speech Recognition (ASR), using the FLEURS benchmark. Our results show a +3.6% F1 score improvement for SLID and a -2.1% average Word Error Rate (WER) reduction for ASR over mHuBERT-147, and demonstrates competitiveness with larger SSL models such as MMS and XEUS. Further analysis shows that ASR models trained on AfriHuBERT exhibit improved cross-corpus generalization and are competitive in extremely low-resource ASR scenarios.