Choose the right engine
Use SIP for infrastructure, Twilio Voice for carrier and Programmable Voice calls, or AI Voice Agent for real conversation testing.
Features
Everything you need to test voice systems—and understand the answer.
Explore what SIPforge does, when to use each test engine, and how to read the result without digging through raw tooling.
How the workflow fits together
From setup to evidence in six clear steps.
Prepare what the test needs
Add servers, accounts, numbers, media, voice-agent prompts, and AI provider keys once, then reuse them safely.
Start with a clean baseline
Begin with low concurrency, short durations, and a clear cost cap. Increase load only after the first result looks correct.
Monitor the run live
Watch active calls, ASR, failures, spend, voice quality, AI latency, and infrastructure resources while the test is moving.
Inspect the evidence
Open call detail for SIP errors, provider status, DTMF, transcripts, callbacks, timing, and per-call cost.
Export or repeat
Share a PDF, export raw CSV or JSON, compare a change, or save the configuration as a reusable template.
Platform capabilities
Every feature supports a decision.
SIPforge keeps traffic generation, live analysis, troubleshooting, and reporting connected.
One testing workspace
Launch and review every voice test from the same history.
- SIP / SIPp load tests
- Twilio Voice call flows
- AI voice-agent conversations
Live test visibility
See whether the system is holding up before the run finishes.
- Call rate and concurrency
- ASR and terminal outcomes
- MOS, jitter, loss, and latency
Failure investigation
Move from a red number to the evidence needed to troubleshoot.
- SIP response details
- Provider and callback errors
- Call logs and transcripts
Infrastructure context
Connect call behavior with what happened on the target server.
- CPU and RAM
- Open file descriptors
- Asterisk channels and drops
Repeatable operations
Turn an ad-hoc test into a dependable quality process.
- Reusable templates
- Scheduled regression runs
- Metric threshold alerts
Team-ready evidence
Give every stakeholder the right level of visibility.
- Admin, runner, and viewer roles
- Run-to-run comparison
- PDF, CSV, and JSON exports
Choose the right engine
Practical guidance before you run.
Open a guide to see what to prepare, how to run the test, and what its results mean.
SIP / SIPp testingGenerate signaling and media traffic against Asterisk, FreeSWITCH, Kamailio, SBCs, proxies, and gateways.
Before you start
- Create a server with host, port, transport, and credentials.
- Confirm firewall, NAT, RTP range, and SIP transport reachability.
How to use it
- Begin with OPTIONS or INVITE/BYE at a low call rate.
- Set total calls, concurrency, hold time, transport, and codec.
- Increase CPS only after a clean baseline.
How to read results
- 200 responses show successful signaling.
- SIP errors identify rejected or timed-out call legs.
- MOS, jitter, loss, and RTT explain media quality.
Twilio Voice testingPlace real Programmable Voice calls and capture status, DTMF, cost, callbacks, and call-flow evidence.
Before you start
- Connect a Twilio account and sync caller numbers.
- Upload media when the flow plays audio and set a first-run cost cap.
How to use it
- Choose a single number or a sequential or random number pool.
- Add destinations and select playback, gather, or DTMF flow.
- Watch callbacks to confirm Twilio reaches SIPforge.
How to read results
- Completed means the call connected and finished.
- Busy and no-answer affect ASR but are not platform failures.
- Failed means Twilio reported a real call failure.
AI voice-agent testingRun real phone conversations, capture transcripts, measure responsiveness, and score whether the agent achieved its goal.
Before you start
- Create a test persona, opening line, goal, and evaluation criteria.
- Connect telephony plus the required LLM, speech-to-text, and text-to-speech providers.
How to use it
- Start with one call to a known test number.
- Listen for pacing, interruptions, and first-response delay.
- Tune the prompt and providers before scaling.
How to read results
- Transcript shows what the system heard and said.
- Latency isolates delays in STT, LLM, TTS, or telephony.
- End reason explains hangups, timeouts, stream loss, or provider errors.
Read the result
The metrics, in plain language.
ASR
Answer-Seizure Ratio
The percentage of call attempts that connected successfully.
MOS
Mean Opinion Score
A 1–5 estimate of call audio quality. Four or higher is excellent.
Jitter
Packet timing variation
Lower is better. High jitter often sounds choppy or robotic.
Packet loss
Audio packets not received
Under 1% is usually healthy; rising loss damages intelligibility.
Latency
Time before a response
Shows where a voice or AI experience begins to feel slow.
Failure
A real engine or provider error
Different from busy or no-answer outcomes, which mainly affect ASR.
Want to see these metrics in a finished test?
Explore the annotated example before creating your first workspace.
From a readiness run
The conclusion, ready to repeat.
Anonymized examples of what a representative SIP and RTP report makes easy to say.
“Sustained the planned call rate with voice quality still in the excellent range.”
“The failure spike sat next to CPU, so the capacity question had an answer.”
“Sign-off used the PDF conclusion. Engineering kept the SIP detail underneath.”
Ready when you are
Know what to test—and what the answer means.
Create your workspace, choose the right engine, and begin with a clear baseline.