mSTEB: Massively Multilingual Evaluation of LLMs on Speech and Text Tasks

TL;DR

This paper introduces mSTEB, a comprehensive benchmark for evaluating multilingual large language models on speech and text tasks across diverse languages, highlighting significant performance gaps for low-resource languages.

Contribution

The paper presents mSTEB, the first standardized benchmark for multilingual LLM evaluation on speech and text, covering low-resource languages and multiple tasks.

Findings

Significant performance gaps for low-resource languages.

High-resource languages outperform low-resource counterparts.

Need for increased focus on under-represented languages.

Abstract

Large Language models (LLMs) have demonstrated impressive performance on a wide range of tasks, including in multimodal settings such as speech. However, their evaluation is often limited to English and a few high-resource languages. For low-resource languages, there is no standardized evaluation benchmark. In this paper, we address this gap by introducing mSTEB, a new benchmark to evaluate the performance of LLMs on a wide range of tasks covering language identification, text classification, question answering, and translation tasks on both speech and text modalities. We evaluated the performance of leading LLMs such as Gemini 2.0 Flash and GPT-4o (Audio) and state-of-the-art open models such as Qwen 2 Audio and Gemma 3 27B. Our evaluation shows a wide gap in performance between high-resource and low-resource languages, especially for languages spoken in Africa and Americas/Oceania. Our findings show that more investment is needed to address their under-representation in LLMs coverage.