Note
Access to this page requires authorization. You can try signing in or changing directories.
Access to this page requires authorization. You can try changing directories.
[This article is prerelease documentation and is subject to change.]
For real-time voice agents, use your voice to test your agent directly in Copilot Studio. This feature lets you speak to your agent and hear its spoken responses, which allows you to quickly validate the voice conversation experience without setting up a phone call.
Note
The test panel only supports interactive voice response agents that use the Standard and Premium voice tier (Realtime voice). Agents that use other voice tiers aren't supported for voice testing in the test panel.
Start a test session
To start a voice test conversation:
Open your agent in Copilot Studio.
Ensure that Realtime voice is selected.
Open the test panel.
In the test panel toolbar, set Chat mode to Speech & DTMF. If Chat mode is currently set to Text, select the Chat mode dropdown in the toolbar, choose Speech & DTMF (Preview), and confirm the mode switch in the dialog. Your browser might prompt you to allow microphone access; if so, select Allow.
Important
Your agent's Security > Authentication setting must be configured to No authentication to use the test with a microphone option.
In the test panel, select **Start voice conversation—the green call button in the center of the panel. The panel shows Connecting… while the voice session is being established.
When connected, the agent speaks its greeting message through your speakers or headphones. The greeting also appears in the transcript as an agent message.
After the agent finishes speaking, speak your response into the microphone. The agent transcribes your speech and displays it as a message in the chat, then responds in both text and voice.
While the agent is speaking, the label Talk to interrupt is shown above the controls. You can start speaking at any time to interrupt the agent (barge-in). When the agent is listening, the label changes to Listening….
Use the control buttons below Talk to interrupt for in-call actions: mute or unmute the microphone, open the DTMF keyboard to send dial-pad input, or select Hide conversation to hide the chat transcript during the call.
To end the current session, select the red End conversation button. The transcript shows Call has ended with a timestamp, and the panel returns to Start voice conversation.
To start a new session, select Start voice conversation again, or select Start new test session in the toolbar.
To switch back to text testing at any time, select the three dots (…) in the test panel title bar, and change Chat mode to Text. Switching modes resets the conversation history. You can download the current history by choosing Switch and download snapshot in the confirmation dialog.
Limitations and differences from real phone calls for IVR agents
Voice testing in the test panel uses the microphone, speakers, or headset of your device. This setup is fundamentally different from the audio hardware and network path used in real phone calls. Be aware of the following differences:
Audio hardware: The test panel captures audio through your device microphone and plays back through your device speakers or headset. Real phone calls use the telephone handset, headset, or speakerphone on the caller's end. These options have different microphone sensitivity, speaker characteristics, and noise cancellation behavior.
Acoustic environment: Your test environment is likely different from the acoustic environment of a real caller. For example, your test environment is probably a quiet office where you use a high-quality headset. A real caller might be in a noisy call center, a car, or a public space with background noise.
Audio path: In the test panel, audio travels directly between your browser and the agent service. In a real phone call, audio passes through the telephony network (PSTN or SIP trunk). This path introduces extra latency, compression, and potential audio quality differences.
Because of these differences, certain voice features and settings might behave differently in the test panel than they do in a real phone call. Use caution when relying on test panel results for the following features:
Barge-in: The barge-in experience depends heavily on the audio hardware and echo cancellation. Barge-in behavior in the test panel might not accurately reflect how it works over a real phone call.
Voice Activity Detection (VAD) settings: VAD sensitivity, end-of-speech timeout, and silence detection thresholds are tuned for telephony audio characteristics. Adjusting these settings based on test panel behavior might lead to suboptimal results in production phone calls.
Speech quality: Background noise, microphone quality, and audio compression all affect voice quality. Test panel results might be more accurate or less accurate than real phone call.
Important
Voice testing in the test panel is useful for quickly validating conversation flow, topic routing, and agent responses. However, it shouldn't replace real phone call testing for tuning voice-specific features like barge-in and VAD settings under real-world telephony conditions. Always perform end-to-end phone call testing before deploying your IVR agent to production.