Scale AI launches Voice Showdown benchmark for voice AI

Scale AI has introduced Voice Showdown, a new benchmark designed to evaluate voice AI models using real-world human interactions rather than synthetic speech. This platform allows users to freely interact with leading AI models and, in exchange for occasional preference votes, gain access to premium services at no cost. The benchmark addresses limitations of existing tools by incorporating diverse languages, accents, background noise, and conversational nuances, providing a more accurate assessment of model performance in practical scenarios. Early results highlight significant capability gaps, particularly in multilingual performance and conversational degradation over time, offering valuable insights into the current state of voice AI.