Quantum Automations Quantum Automations
Blog · Portfolio
← Back to Blog
Guide · Voice AI

Voicemail Detection Settings That Actually Work

Published May 2026
Topic Voice AI · Outbound
Reading time 5 min
For UK SMEs
On this page
  1. Recommended Baselines
  2. QA Checklist
  3. Good / Bad / Ugly

Recommended Baselines

  • Greeting length threshold: 1.6–2.4s (longer → voicemail likelihood)
  • Beep detection: enable with 500–1200Hz band emphasis and -28dB threshold
  • Silence window: 600–900ms before first agent utterance
  • DTMF presence: deprioritize voicemail if early DTMF detected
  • Twilio AMD: use DetectMessageEnd with 2.0s max greeting

Hybrid Heuristic

// Pseudocode combining AMD + custom VAD/beep
const decision = (amd, vad, beep, dtmf, tGreeting) => {
  if (beep.detected && tGreeting > 1.7) return 'VOICEMAIL';
  if (amd.result === 'MACHINE' && vad.noHumanSpeech && tGreeting > 1.8) return 'VOICEMAIL';
  if (dtmf.early) return 'HUMAN';
  if (vad.humanSpeechStart < 1200) return 'HUMAN';
  return amd.result === 'MACHINE' ? 'VOICEMAIL' : 'HUMAN';
};

QA Checklist

  • Measure FP/FN rates against 100+ labeled calls; track per-carrier
  • Record features (greeting length, beep time, VAD events) for audits
  • A/B AMD settings monthly; carriers change cadence
  • Fail-safe: if uncertain → short intro + re-check before full pitch

Good / Bad / Ugly

  • Good: Beep detection + AMD + VAD fusion; measurable thresholds.
  • Bad: Relying on one signal (AMD only) → drift over time.
  • Ugly: Double beeps, carrier ringback fakeouts; use cooldown + retry strategy.

Related Reading

Voice AI Architecture 2025

Production blueprint for AI voice agents

Appointment Playbook 2025

Booking optimization strategies

Wrestling with AMD reliability?

30-minute audit. We map your stack, your constraints, and where AI will pay back fastest.

Take the Quantum Leap →
© 2026 Quantum Automations Group Ltd
Home Blog Portfolio Privacy Terms Security