Researchers have introduced SPEARBench, a novel benchmark for evaluating the naturalness of streaming speech-to-speech language models in conversational settings1. This benchmark addresses the limitations of existing speech and text benchmarks, which fail to capture the nuances of human-like interactions, including timing, turn-taking, and prosody. SPEARBench assesses the ability of these models to engage in natural-sounding conversations, taking into account factors such as language and dialect consistency, interpersonal stance, and relationship-aware appropriateness. By using this benchmark, developers can improve the performance of their speech-to-speech models, making them more effective in real-world applications. The introduction of SPEARBench has significant implications for the development of more sophisticated and human-like language models, which can be used in a variety of applications, from customer service to language translation. This matters to practitioners because it enables them to create more realistic and engaging conversational interfaces, ultimately enhancing user experience.