Real-time voice agents need new tests for timing and outcomes
Evaluating Real-Time Voice Agents: From Component Quality to Grounded Outcomes
Artificial Intelligence
Summary
Real-time voice assistants are tested differently by engineers, linguists, and evaluators, making it hard to judge if one is really better. The authors found that how an assistant is built is less important than how well it recovers from interruptions and proves it did its job. They also highlight that conversations with more than two people need new ways to measure when and what the assistant should say. They suggest a new standard to report voice agent performance based on timing, recovery from disruptions, and verified results.
What this means in practice
- •For voice assistant developers: Use the TRG standard to measure timing, recovery, and verified outcomes for improving real-time voice agents beyond existing benchmarks focused on components.
- •For customer service platform teams: Integrate multi-speaker reasoning and turn-taking benchmarks to better handle conversations with multiple participants and improve agent effectiveness.
A survey. It maps existing work.
Authors
Shivam Negi, Arpit Rawat, Rashi Jain
Abstract
Real-time voice agents have moved from research prototypes to production deployments, yet the literature describing them is fragmented across three communities that rarely cite one another: speech foundation modelling, turn-taking psycholinguistics, and agentic evaluation. Architecture papers report latency, turn-taking papers report prediction accuracy, and agentic benchmarks report task success, so no single number describes whether a deployed agent is actually good. We address that gap with three evidence-based claims, each traceable to a corpus of 38 primary sources organised into an application-centric taxonomy of six categories. First, architecture choice is a deployment constraint rather than a settled verdict: a 2026 enterprise tutorial reports that no fully self-hostable end-to-end system yet meets production constraints, while a chunked cascade independently reaches state-of-the-art duplex behaviour, showing duplex behaviour is separable from duplex architecture. Second, evaluation has shifted decisively from component quality toward grounded outcomes, with recent benchmarks verifying backend state rather than trusting what the agent claims to have done. Third, the dyadic assumption in most models and benchmarks is breaking down: multiparty turn-taking and multi-speaker reasoning benchmarks show that deciding when not to speak, and reasoning about who may be told what, are first-class capabilities two-participant framings cannot measure. For each source we state the problem it targets, its mechanism, and its reported evidence, alongside the search strategy, inclusion criteria, and a verification step that caught a misattributed arXiv identifier in circulation. We propose TRG (Timing-Recovery-Grounded), a reporting standard characterising an agent by timing, post-disruption recovery, and state-verified outcome together, with a conditional fourth axis for multiparty deployments.