NaijaS2ST - Speech Translation to English (Zero-Shot): leaderboard

Metric: SSA-COMET (0-100, COMET for African languages) of speech-to-text translation on the NaijaS2ST test set, Nigerian-language speech to English text, mean of Hausa, Igbo, Nigerian Pidgin and Yoruba; audio LLM prompted zero-shot; higher is better. Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 4 models tracked.

Top models

#ModelScore
1NaijaS2ST Gemini 3.1 (checkpoint unspecified)62.1
2Gemini 2.551.6
3GPT-Audio40.4
4Gemma 440.2

Interactive version: theaggregate.ai/benchmark?slug=naijas2st-speech-translation-to-english-zero-shot · How It Works · Data refreshed daily, snapshot 2026-10-07.