Vox-Infinity benchmark evaluates long-context spoken language models

Vox-Infinity is a new benchmark designed to measure the limits of long-context understanding in speech models. It evaluates how models maintain semantic and acoustic cues across increasing audio turn counts and durations.