AI AMD
Answering Machine Detection powered by real-time speech models.
Legacy AMD uses energy detection and silence patterns — it works in a quiet lab and fails in the real world. Our AI AMD uses real-time speech models fine-tuned on call center audio to distinguish a live human from an answering machine greeting within 1.5 seconds, with false-positive rates under 5% on live calls. It runs as a FreeSWITCH or Asterisk module and drops into your existing Vicidial or custom dialer without a platform migration. We also offer domain adaptation — if your campaign calls a specific demographic or geography where stock models underperform, we retrain on your data.
Features
Real-time speech classification
Models run at the media layer, not post-call. Detection fires within 1-2 seconds of connect — before agents are bridged to machines.
Low false-positive rate
Under 5% on live calls in standard deployments. Custom model training reduces this further for specific demographics or geographies.
Drop-in for FreeSWITCH and Asterisk
Ships as a FreeSWITCH app and an Asterisk AGI/ARI module. Dialplan changes take less than a day to implement.
Custom model training
Send us a sample of your calls. We retrain the model on your audio and validate before deploying. Typical improvement: 20-40% reduction in false positives.
Webhook events
Human, machine, and silence events delivered via webhook with timestamp and confidence score. Connect to your CRM or dialer logic.
Audit and analytics
Per-campaign false-positive rates, detection latency, and machine vs. human breakdowns — not just a binary event log.
Pricing
Starting at $0.002/call analyzed. Custom pricing for high-volume deployments.
Common questions
Related products & services
Ready to build on carrier-grade voice?
Talk to a VoIP engineer — not a salesperson.