What to Evaluate
When listening to a test call, focus on the following areas:Pronunciation
| Check | Common Issues |
|---|---|
| Proper nouns (company names, person names) | Mispronounced due to unusual spelling |
| Numbers and currency amounts | ”One thousand two hundred” vs. “twelve hundred” |
| Dates and times | ”June third” vs. “June three” |
| Acronyms | Spoken letter-by-letter vs. as a word |
| Domain-specific terms | Industry jargon the TTS engine may not recognize |
Response Timing
- How long does it take for the agent to respond after the caller stops speaking?
- Does the agent begin responding before the caller finishes (false triggers)?
- Is there a noticeable pause between the caller’s input and the agent’s response?
Voice Quality
- Is the voice clear and intelligible?
- Does background noise affect the audio (test from a noisy environment to assess denoising)?
- Does the voice sound natural and appropriate for the use case?
- Is the speech rate comfortable for the expected caller audience?
End-Call Behavior
- Does the agent end the call correctly when the conversation concludes?
- Does the silence timeout trigger appropriately when there is no caller input?
- Is the goodbye message natural and complete before the call ends?
Place a Test Call
Click Test Call
Click the Test Call button in the agent detail view. A dialog opens asking for your phone number.
Enter your phone number
Enter the phone number to call in E.164 format (e.g.,
+14155550100). DialNexa will call this number. Answer the call to begin the test session.Conduct the test
Speak with the agent as a caller would. Work through your key scenarios. Pay attention to the audio quality items listed above.
Test calls are real phone calls routed through the telephony stack. They consume concurrency slots and are billed at your standard per-minute rate. Keep test calls focused to manage costs.
Review the Call Record
After the call:- Open Call History and find the test call.
- Listen to the call recording if recording is enabled.
- Review the transcript for transcription accuracy. Transcription errors indicate words the ASR engine struggles with, which can cause the LLM to receive incorrect input.
- Check the call duration and timeline to identify timing issues.
Annotate Test Results
In the call detail view, you can add notes to document issues found during testing:- Click Add Note in the call detail view.
- Describe the issue: what you expected, what happened, and the timestamp in the recording where the issue occurs.
- Save the note. It is attached to the call record for reference during debugging.
Fix Common Audio Issues
| Issue | Where to Fix |
|---|---|
| Mispronounced word | Agent Setup: Audio Tab. Add a custom pronunciation in the voice model settings, or rephrase the text in the prompt to use a phonetic spelling. |
| Slow response timing | Agent Setup: Engine Tab. Increase response eagerness or enable predictive preprocessing. |
| Speech too fast or slow | Agent Setup: Audio Tab. Adjust the Speed slider. |
| Voice sounds robotic or unnatural | Agent Setup: Audio Tab. Try a different voice model or adjust Stability. |
| Background noise on caller’s end audible | Agent Setup: Audio Tab. Enable Denoising. |
| Agent does not hang up properly | Agent Setup: Call Tab. Review silence timeout and max call duration settings. |