# SunflowerASR sets new error-rate record for African language speech recognition

> A model trained with Group Relative Policy Optimisation beats prior results on two African speech benchmarks.

Oossa · 2026-10-23 · https://oossa.com/en/sunflowerasr-sets-new-error-rate-record-for-african-language-speech-recognition

Researchers released SunflowerASR, an encoder‑decoder speech recogniser that uses Group Relative Policy Optimisation (GRPO) to reduce errors. The model builds on Whisper‑large‑v3 and was tested on 51 African languages. It achieves the lowest published macro‑averaged word error rate (WER) on the AfriVox‑v2 benchmark (0.227) and on SimbaBench (0.372). The gains are strongest when training data are limited or audio is noisy.

## The facts

- Paper posted on arXiv on 2026-10-23
- SunflowerASR WER: 0.227 on AfriVox‑v2, 0.372 on SimbaBench

## Why it matters

The improvement helps developers create more reliable voice tools for speakers of low‑resource African languages.

## Sources & references

1. [Group Relative Policy Optimisation Improves Multilingual Speech Recognition in Low-Resource Languages](https://arxiv.org/abs/2610.30663) – arXiv, 2026-10-23

Last updated: 2026-10-23