AWS ML BlogTuesday · June 9, 2026FREEEvaluate your Amazon Nova Sonic voice agent at scale, no microphone required
The Nova Sonic Test Harness, introduced in an AWS ML Blog post, is an open-source framework designed to address two key challenges in voice agent development: rapid iteration and comprehensive evaluation. It runs complete multi-turn conversations with Amazon Nova Sonic automatically, eliminating the need for a microphone. The harness uses LLM-as-judge techniques to evaluate agent quality at scale and can detect audio hallucinations—cases where the model's audio output doesn't match its text output. This allows developers to tune system prompts and tool configurations quickly by running a conversation, seeing results, adjusting, and repeating. The framework serves both as a rapid iteration tool and as a validation framework for production-scale testing, making it easier to build reliable voice agents without manual testing overhead.
Developers can now test voice agents at scale without hardware, speeding up iteration and catching audio hallucinations.