How VICIdial decides a call is a machine, not a person
VICIdial's AMD listens to initial silence, greeting length, and the pause after the greeting to guess human vs machine — and long greetings fool it.
VICIdial decides a call is a machine by listening to the audio right after the call is answered and measuring a few timing patterns: how long the line is silent before anyone speaks, how long the greeting runs, and how long the silence after the greeting lasts. A human says "hello?" and then stops; an answering machine plays a longer, scripted greeting and then beeps.
What the analysis listens for
When a call connects, the dialer runs an analysis application over the first few seconds before it hands the call to anyone. That window is where AMD (answering machine detection) makes its call. It is watching three things in particular:
- Initial silence — the quiet between the line answering and the first sound. Machines often have a longer lead-in.
- Greeting length — how long the first burst of speech runs. "Hi" is short; "You've reached the voicemail of..." is long.
- After-greeting silence — the pause once the speech stops. A person pauses briefly and waits; a machine pauses, then beeps.
All of this happens inside a short, fixed analysis budget. The application has a total time it is allowed to listen before it must return a verdict, so it can't sit there forever waiting for a beep — it weighs what it has heard against its thresholds and decides.
flowchart TD
A[Call answered] --> B[Measure initial silence]
B --> C[Measure greeting length]
C --> D[Measure after-greeting silence]
D --> E{Within human thresholds?}
E -->|Yes| F[Human]
E -->|No, long greeting or beep| G[Machine]Where beep detection fits
On top of the timing pattern, the analysis also does Beep detection. A clean tone after a long greeting is a strong signal that you are on voicemail and that the moment to leave a message has arrived. Together, greeting timing plus the beep give a more confident machine verdict than either signal alone.
Why long greetings cause false positives
The weakness is built into the method. The whole thing leans on "machines talk longer than people." So a chatty human who answers with "Hello, this is the Johnson residence, who's calling?" can blow past the greeting-length threshold and get tagged as a machine. That is an AMD false positive: a real person who gets hung up on.
This is also why machine answers carry their own Status (lead status). Watching that machine rate against your overall Contact rate is how you tell whether AMD is helping or whether it is hanging up on customers.
Knowing the mechanism is half the job; the other half is deciding whether to run it at all. See when to turn AMD on and when to leave it off, and for the full picture read the complete guide to VICIdial AMD and CPD.
Dialing in AMD thresholds means editing your dialplan and watching the results live. VICIfast hands you a dedicated VICIdial server, ready in under 40 seconds, so you can tune it without sharing a box. See plans and pricing.
About VICIfast LLC
VICIfast LLC operates a managed VICIdial hosting + BYOI service for outbound and inbound call centers. We run the dialers, the carriers, the recordings pipeline, and the compliance plumbing so operators don’t have to.
Citing this article
VICIfast Engineering. “How VICIdial decides a call is a machine, not a person”. VICIfast LLC, June 24, 2026. Retrieved from https://vicifast.com/blog/how-vicidial-amd-detects-machines
Have questions?
Related posts
You might be interested in
VICIfast newsletter
Liked this? Get the next one in your inbox.
We ship the kind of stuff you just read — concrete, numbers-first, no drip. One email when a new post goes live. Unsubscribe in one click.
Comments
No comments yet — be the first.