τ-Voice Benchmark

A speech agent evaluation framework designed by a Stanford University research team, focused on measuring a model's ability to complete end-to-end tasks in realistic multi-turn conversational settings. It encompasses speech understanding, tool invocation, and multi-step planning.