Add dataset card for Vox-Infinity

README.md

@@ -0,0 +1,12 @@
+---
+task_categories:
+- audio-text-to-text
+---
+
+# Vox-Infinity
+
+This repository contains the Vox-Infinity dataset from the paper [Vox-Infinity: Benchmarking the Limits of Long-Context Spoken Language Models](https://huggingface.co/papers/2609.22452).
+
+Project page: https://vox-infinity.github.io
+
+Vox-Infinity is the first benchmark designed to evaluate long-context understanding in spoken language models. It systematically extends audio history along two dimensions: turn count and turn duration, covering diverse scenarios with varying interaction structures and semantic complexity. It provides explicit answer-provenance annotations and organizes samples according to the amount of historical context required to resolve each query.