Start by calling an echo test line, which plays your voice input back to your handset. (So you hear yourself again some time later). This is "round trip" latency.
I used 909 390 0003.
The best analysis method is to record the audio in a program that displays audio in a graphic format, plotted against time. (The free Audacity is frequently suggested.)
On the graph, you can note the time or the original signal against the echo. See screenshots in this post from the incredible VoIPFanRP on the RingPlus forum, who was my go-to source on testing and understanding latency:
https://social.ringplus.net/discussion/comment/78438/#Comment_78438
However, I used a far simpler measuring method that only requires a stopwatch, and the ability to "keep time" by turning the echo line response into an impromptu metronome.
For input, use a short plosive word. (I use "check"). Call the echo line, say "check". Listen for the echo "check". Say "check" again. Repeat until you can get a predictable metronome like rhythm, with every other "check" being produced by you, and the intervening "check" produced by the echo.
When you have a predictable check...check....check....check.... pattern, start the stopwatch when you say check, and stop on the echo check. This is round-trip latency, so divide by 2 to get 1-way latency.
Sounds crude, but it's the same method used by musicians to set tempos to play together, and is quite effective with experience. In repeated tests, I'm usually within 1-2%, or about a 5 ms variance on a 500ms round-trip latency.
For comparisons to really be meaningful, you need to limit the variables as much as possible. My tests were done using identical phones, and repeated several times & averaged. You also need to have a baseline for comparison-- in my case, the same phone on a direct-cellular connection. That baseline figure is then subtracted from the latency of any particular test of another voice path.
(Of interest: in the VoIPFanRP post linked above, he saw 200 ms variance in latency between 2 different model phones via the same path.)