SpeechR: A Benchmark for Speech Reasoning in Large Audio-Language Models

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Yang, Wanqi, Li, Yanda, Wei, Yunchao, Fang, Meng, Chen, Ling
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866915425196965888
author Yang, Wanqi
Li, Yanda
Wei, Yunchao
Fang, Meng
Chen, Ling
author_facet Yang, Wanqi
Li, Yanda
Wei, Yunchao
Fang, Meng
Chen, Ling
contents Large audio-language models (LALMs) have achieved near-human performance in sentence-level transcription and emotion recognition. However, existing evaluations focus mainly on surface-level perception, leaving the capacity of models for contextual and inference-driven reasoning in speech-based scenarios insufficiently examined. To address this gap, we introduce SpeechR, a unified benchmark for evaluating reasoning over speech in large audio-language models. SpeechR evaluates models along three key dimensions: factual retrieval, procedural inference, and normative judgment. It includes three distinct evaluation formats. The multiple-choice version measures answer selection accuracy. The generative version assesses the coherence and logical consistency of reasoning chains. The acoustic-feature version investigates whether variations in stress and emotion affect reasoning performance. Evaluations on eleven state-of-the-art LALMs reveal that high transcription accuracy does not translate into strong reasoning capabilities. SpeechR establishes a structured benchmark for evaluating reasoning in spoken language, enabling more targeted analysis of model capabilities across diverse dialogue-based tasks.
format Preprint
id arxiv_https___arxiv_org_abs_2508_02018
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle SpeechR: A Benchmark for Speech Reasoning in Large Audio-Language Models
Yang, Wanqi
Li, Yanda
Wei, Yunchao
Fang, Meng
Chen, Ling
Computation and Language
Artificial Intelligence
Large audio-language models (LALMs) have achieved near-human performance in sentence-level transcription and emotion recognition. However, existing evaluations focus mainly on surface-level perception, leaving the capacity of models for contextual and inference-driven reasoning in speech-based scenarios insufficiently examined. To address this gap, we introduce SpeechR, a unified benchmark for evaluating reasoning over speech in large audio-language models. SpeechR evaluates models along three key dimensions: factual retrieval, procedural inference, and normative judgment. It includes three distinct evaluation formats. The multiple-choice version measures answer selection accuracy. The generative version assesses the coherence and logical consistency of reasoning chains. The acoustic-feature version investigates whether variations in stress and emotion affect reasoning performance. Evaluations on eleven state-of-the-art LALMs reveal that high transcription accuracy does not translate into strong reasoning capabilities. SpeechR establishes a structured benchmark for evaluating reasoning in spoken language, enabling more targeted analysis of model capabilities across diverse dialogue-based tasks.
title SpeechR: A Benchmark for Speech Reasoning in Large Audio-Language Models
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2508.02018