# How Architecture, Typology, and Adaptation Strategies Shape ASR for Low-Resource African Languages: A Controlled Empirical Study

**Authors:** Mercy Muchai, Kevin Chege, Nick Mumero Mwangi, Samuel Chege Maina, Felermino Ali, Elizabeth Ankrah, Ignatius Ezeani  
**Lab:** Microsoft · **Published:** 28 Oct 2026  
**Links:** [arXiv](https://arxiv.org/abs/2610.39004) · [PDF](https://arxiv.org/pdf/2610.39004) · [This page](https://omen-bit.github.io/ai-paper-daily/paper/2610.39004.html)

## Abstract

Despite recent advancements in ASR, identifying effective architectures, fine-tuning approaches, and data adaptation strategies for low-resource languages remains challenging. We present a systematic comparison across five low-resource African languages and Swahili as a mid-resource contrast, fine-tuning three models from distinct architectures: Phi-4-multimodal-instruct, Whisper-large-v3-turbo, and MMS-1B-all. We analyze data sufficiency, monolingual versus multilingual adaptation, domain fine-tuning, and language-adapter transfer. Phi-4 achieves the best overall accuracy, MMS shows comparable CER, while Whisper exhibits high CER, consistent with known hallucinations and repetition loops. Performance saturates at different data scales, offering practical data-efficiency targets, while typological complexity consistently shapes error rates. Domain fine-tuning benefits Phi-4 and Whisper but degrades MMS, revealing limitations of adapter-based adaptation. Proxy language adapters outperform training MMS adapters from scratch.

---

Notes written by AI from the paper, numbers copied from the abstract. Check the paper before you rely on them. From AI Papers Daily.