SPEARBench: A Benchmark for Naturalness Evaluation in Streaming Speech-to-Speech Language Models

2026-07-06

AI Summary

Existing speech and text benchmarks inadequately assess the naturalness of streaming speech-to-speech language models in conversational settings. This work proposes SPEARBench, the first multidimensional evaluation framework tailored to such models, capturing core characteristics of human dialogue—including response latency, turn-taking dynamics, emotional naturalness, interpersonal stance, and dialect consistency. Built upon the Seamless Interaction corpus, the benchmark employs controlled dialogue prompts, integrates multi-model inference, ASR robustness testing, and emotion-and-stance analysis, and introduces interpretable distributional baselines for systematic evaluation. Experimental results reveal that while current models achieve high signal quality and low ASR error rates, they still exhibit significant deviations from human-like behavior in temporal interaction patterns and sociolinguistic dynamics.

Abstract

Streaming speech-to-speech language models aim to answer spoken queries directly with synthetic speech. However, standard speech and text benchmarks do not capture whether these systems behave naturally in conversations, where timing, turn-taking, prosody, interpersonal stance, language and dialect consistency, and relationship-aware appropriateness jointly shape perceived quality. We introduce SPEARBench, a benchmark for evaluating naturalness in speech-to-speech language models from question-answer interactions. SPEARBench constructs controlled dialogue prompts from the Seamless Interaction corpus, runs inference across multiple models, and evaluates generated answers using a multidimensional protocol that covers response latency, interruptions, speech quality, ASR robustness, language and dialect consistency, emotional naturalness, interpersonal stance, and explainable distributional baselines. The benchmark includes original human answers as a reference condition and reports results for several contemporary models. Results show that current models can achieve high signal-level quality and low ASR error while still differing from human conversational behavior in latency, overlap, dialect preservation, emotional adaptation, and interpersonal stance dynamics.

Problem

• naturalness
• speech-to-speech
• conversational behavior
• benchmark
• streaming

Innovation

• speech-to-speech language models
• naturalness evaluation
• conversational benchmark
• multidimensional assessment
• human-like interaction