Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Ethio-ASR: Joint Multilingual Speech Recognition and Language Identification for Ethiopian Languages

Badr M. Abdullah, Israel Abebe Azime, Atnafu Lambebo Tonja, Jesujoba O. Alabi, Abel Mulat Alemu, Eyob G. Hagos, Bontu Fufa Balcha, Mulubrhan A. Nerea, Debela Desalegn Yadeta, Dagnachew Mekonnen Marilign, Amanuel Temesgen Fentahun, Tadesse Kebede, Israel D. Gebru, Michael Melese Woldeyohannis, Walelign Tewabe Sewunetie, Bernd Möbius, Dietrich Klakow

Abstract

We present Ethio-ASR, a suite of multilingual CTC-based automatic speech recognition (ASR) models jointly trained on five Ethiopian languages: Amharic, Tigrinya, Oromo, Sidaama, and Wolaytta. These languages belong to the Semitic, Cushitic, and Omotic branches of the Afroasiatic family, and remain severely underrepresented in speech technology despite being spoken by the vast majority of Ethiopia's population. We train our models on the recently released WAXAL corpus using several pre-trained speech encoders and evaluate against strong multilingual baselines, including OmniASR. Our best model achieves an average WER of 30.48% on the WAXAL test set, outperforming the best OmniASR model with substantially fewer parameters. We further provide a comprehensive analysis of gender bias, the contribution of vowel length and consonant gemination to ASR errors, and the training dynamics of multilingual CTC models. Our models and codebase are publicly available to the research community.

Ethio-ASR: Joint Multilingual Speech Recognition and Language Identification for Ethiopian Languages

Abstract

We present Ethio-ASR, a suite of multilingual CTC-based automatic speech recognition (ASR) models jointly trained on five Ethiopian languages: Amharic, Tigrinya, Oromo, Sidaama, and Wolaytta. These languages belong to the Semitic, Cushitic, and Omotic branches of the Afroasiatic family, and remain severely underrepresented in speech technology despite being spoken by the vast majority of Ethiopia's population. We train our models on the recently released WAXAL corpus using several pre-trained speech encoders and evaluate against strong multilingual baselines, including OmniASR. Our best model achieves an average WER of 30.48% on the WAXAL test set, outperforming the best OmniASR model with substantially fewer parameters. We further provide a comprehensive analysis of gender bias, the contribution of vowel length and consonant gemination to ASR errors, and the training dynamics of multilingual CTC models. Our models and codebase are publicly available to the research community.
Paper Structure (29 sections, 4 figures, 7 tables)

This paper contains 29 sections, 4 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Ethiopian Languages: An Overview
  3. Amharic
  4. Oromo
  5. Tigrinya
  6. Sidaama
  7. Wolaytta
  8. Writing Systems
  9. Challenges for ASR Systems
  10. Dataset
  11. WAXAL Dataset
  12. Data-driven Lexical Analysis
  13. Joint ASR and LID Modeling
  14. Model Architecture
  15. Grapheme-based Vocabulary
  16. ...and 14 more sections

Figures (4)

  • Figure 1: Average WER (%) versus model size, where lower WER indicates better performance. Our Ethio-ASR models consistently achieve lower WER across all sizes, outperforming CTC and LLM-based OmniASR baselines.
  • Figure 2: Top: Vocabulary growth curves for Ethiopian languages (WAXAL) and English/French (Multilingual LibriSpeech) across corpus sizes up to 1.4M tokens. Bottom: Type-token ratio (TTR) at 800k tokens.
  • Figure 3: Training dynamics of Ethio-ASR with w2v-BERT-2.0 and MMS-1B as pre-trained models. This analysis reveals that both models learn to transcribe Latin-based scripts before Ge'ez script.
  • Figure 4: Effect of vowel length and geminate normalization on WER across Oromo, Sidaama, and Wolaytta. Results are shown for four conditions: no normalization, vowel length normalization, geminate normalization, and both normalized.