Test voice agents at scale with simulated conversations
Run tests with datasets containing multiple scenarios for your voice agent to evaluate performance across different situations.Create a dataset for testing
Configure your agent dataset template with:
- Agent scenarios: Define specific situations for testing (e.g., “Update address”, “Order an iPhone”)
-
Expected steps: List the actions and responses you expect
Set up the test run
- Navigate to your voice agent and click Test
- Simulated session mode will be pre-selected (voice agents can’t be tested in single-turn mode)
- Select your agent dataset from the dropdown
- Choose relevant evaluators
Only built-in evaluators are currently supported for voice simulation runs. Custom evaluators will be available soon.
Trigger the test run
Click Trigger test run to start. The system will call your voice agent and simulate conversations for each scenario.
Review results
Each session runs end-to-end for thorough evaluation:
- View detailed results for every scenario
- Text-based evaluators assess turn-by-turn call transcription
-
Audio-based evaluators analyze the call recording
Inspect individual entries
Click any entry to see detailed results for that specific scenario.By default, test runs evaluate these performance metrics from the recording audio file:
- Avg latency: How long the agent took to respond
- Talk ratio: Agent talk time compared to simulation agent talk time
- Avg pitch: The average pitch of the agent’s responses
-
Words per minute: The agent’s speech rate
