Search papers, labs, and topics across Lattice.
This paper introduces SPEARBench, a novel benchmark designed to evaluate the naturalness of streaming speech-to-speech language models in conversational contexts. By employing a multidimensional evaluation protocol that assesses factors such as response latency, emotional naturalness, and interpersonal stance, the authors reveal that existing models, despite achieving high signal-level quality, still fall short of replicating human-like conversational behaviors. The findings highlight significant gaps in areas like latency and emotional adaptation, underscoring the need for improved model training to enhance conversational realism.
Current speech-to-speech models may sound good, but they miss the mark on natural conversational dynamics, revealing critical areas for improvement.
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.
Amazon Science