Frequently asked questions about voice load testing.

SIPforge runs repeatable SIP, Twilio Voice, and AI voice-agent tests, then shows live quality metrics, call-level evidence, reports, and trends so a team can decide whether a voice system is ready.

Choose SIP when you are testing your own SIP infrastructure. Choose Twilio Voice when you need real Programmable Voice or PSTN calls. Choose AI Voice Agent when you need to validate a live conversation, including transcript, latency, and whether the agent met its goal.

Run one small test first: one SIP call, one Twilio destination, or one AI voice-agent call. Confirm status, audio, callbacks, cost, and the result before you raise call rate or concurrency.

Viewers can inspect results. Runners and organization admins can launch tests. Screens such as AI providers, agents, and workspace settings stay limited to organization admins.

ASR is answer seizure ratio. In SIPforge it is completed calls divided by total attempts. Busy, no-answer, canceled, and failed attempts stay outside the completed numerator.

A test should fail when a call status is failed or the engine itself errors. Busy, no-answer, and canceled mean the destination was unavailable or the call did not connect. Those outcomes should finish the test and lower ASR instead of marking the run failed.

Busy means the destination was reachable but could not take the call. No-answer means it did not answer before the timeout. Canceled means the call ended before it completed. Failed means Twilio, SIP, or the runner reported a real failure. Only the failed status counts as a call failure.

You need a server record, SIP host, transport, source extension or credentials, a destination, a scenario, a call rate, concurrency, and either a total-call limit or a duration limit.

Those quality metrics appear for SIP scenarios that capture RTP media. A signaling-only scenario may show SIP responses without media-quality scores.

Connected mode uses more than one endpoint credential so the test can model both the caller and the callee, instead of driving only one side of the call.

Twilio tests place real calls on your account. A cost cap stops SIPforge from starting new billable calls after spend reaches the threshold you set.

Confirm the callback URL is public and reachable, the Twilio credentials are valid, and the webhook is responding while the call is still active. Then open the call detail for the exact Twilio event and error text.

Play-and-gather tests play audio and collect digits from the far end. Send-DTMF tests send digits into the call and record how the provider responds.

It places or receives a real phone call, speaks with text-to-speech, listens with speech-to-text, uses a language model to respond, and scores the conversation against the criteria you set.

Maximum call duration is a ceiling, not a guaranteed length. A call can end earlier because the far end hung up, the destination was unavailable, the goal completed, a no-audio timeout fired, the media stream dropped, or a provider returned a terminal status.

High latency makes a live call feel unnatural. SIPforge tracks the time from callee speech to agent audio and can separate delay across speech-to-text, the language model, text-to-speech, and telephony.

No. A template is a saved setup for a new run. Each historical test keeps its own result and configuration snapshot.

Schedule a template that has already passed a manual run. Recurring schedules are for health checks, regression coverage, and quality monitoring, not for the first experiment.

Use PDF for a readable summary, CSV for spreadsheet analysis, and JSON when engineering needs the full machine-readable result.