Skip to content
View original post on X: Design ArenaOfficial· 23/100AI score23/100

Design Arena launches AudioAgentBench to rank speech-to-speech voice agents

AISummary

Design Arena has launched AudioAgentBench, a leaderboard that tests speech-to-speech models as voice agents across six multi-turn, customer-facing scenarios such as booking appointments, placing grocery orders and planning events. The benchmark scores tool use, instruction following and state tracking, using recordings of real human speakers rather than synthetic voices.

Post on XView on X
Design ArenaVerified on X
@DesignArena

Part of a thread · earlier post

AudioAgentBench is a custom Design Arena leaderboard which tests speech-to-speech models as real voice agents across 6 multi-turn, customer-facing scenarios, like booking appointments, placing grocery orders and planning events.

It scores tool use, instruction following and state tracking. Rankings use Real Audio: recordings of real human speakers, not synthetic voices.

Check out the full leaderboard here: https://www.designarena.ai/leaderboard?tab=audio&section=audio

Source: Design Arena · x.comPublished · added here

Topics

Evals & benchmarks

Tags

#Other