Compare answering machine detection options
There are four ways to compare answering machine detection for an outbound dialer: Asterisk's built-in AMD(), transcript-based AMD that turns speech into text, carrier AMD from calling platforms like Twilio, Telnyx and Plivo, and acoustic AI AMD like nomoVM. Each makes different trade-offs.
| Asterisk AMD() | Transcript AMD | Carrier AMD | Acoustic AI (nomoVM) | |
|---|---|---|---|---|
| How it decides | Silence and word timing | Speech-to-text, then words like "leave a message" | The carrier's own detector on its calls | The sound of the answer, plus tone and recording checks |
| Works with any carrier | Yes | Usually | Only calls placed through that carrier | Yes |
| FreeSWITCH | Not available (Asterisk only) | Varies | Through the carrier's API | Lua script + mod_audio_stream |
| Fits VICIdial as is | Built in | Varies | Needs the carrier's API or SIP features | Same AMDSTATUS variable |
| Time to decide | Configurable, often 2 to 5 s | Needs enough words to transcribe | Varies by product and mode | About 2 s, up to 3.2 s |
| Handles "Hello? ... Hello?" | Often as machine | Usually | Varies | Yes |
| Recorded "Hello?" robots | No | Hard: the words are human | Varies | Fingerprint check |
| Cost | Free | Per minute of transcription | Per call, on top of carrier minutes | Free to 30,000 calls, then from $99 a month |
"Varies" means it depends on the specific product and settings. Check each vendor's current documentation for their numbers.
Asterisk AMD()
Free, built into every Asterisk, and what VICIdial uses by default. Good enough on clean lines with typical greetings; weak on real-world answers, because timing alone can't tell a hesitant person from a short greeting. Asterisk AMD alternative · amd.conf settings explained
Transcript-based AMD
Speech-to-text, then look for voicemail phrases. It understands words well, but it needs enough speech to transcribe, which costs time, and a recorded "Hello?" says exactly what a person says. In our own comparison on one outbound day, a hosted transcript AMD kept about 80% of live people; nomoVM kept 94.9% on the same kind of traffic. Details
Carrier AMD
Calling platforms such as Twilio, Telnyx and Plivo offer answering machine detection on calls you place through their APIs. That is convenient if all your calls already go through them, but it ties detection to one carrier and usually doesn't plug into a VICIdial or Asterisk dialplan that uses other trunks. See Twilio, Telnyx and Plivo.
Acoustic AI AMD
A model trained on real answered calls hears the difference between a person and a recording, without waiting for words. nomoVM adds tone detection, a fingerprint check for repeated recordings and a longer listening window for unclear calls. How it works
How to choose
- Measure both errors on your own calls: live people dropped, and machines sent to agents. A single "accuracy" number hides one of them. Our testing guide shows how.
- Check the decision time. Every second of AMD is a second of silence for the person who answered.
- Check what happens when it fails. A detector that times out should hand the call back, not drop it.
- Check the fit. The best detector is the one that works with the dialer and carriers you already have.
Stop sending voicemail to your agents
30,000 calls a month free. No card, no setup fee. Your dialer falls back to Asterisk AMD() on its own if anything goes wrong.