VoiceDetect AMD

AI-Powered Answering Machine Detection

Detect human vs voicemail/IVR in real-time. Make smarter routing decisions in ~3 seconds using speech-to-text and ML classification.

Get in Touch

$ voicedetect --analyze incoming_call.wav

Analyzing audio stream...

Speech detected: "Hi, you've reached John's voicemail..."

Classification: MACHINE (confidence: 94.2%)

Decision: HANGUP (latency: 2.8s)

Features

Real-time Detection

WebSocket streaming for low-latency decisions. Get human/machine classification in approximately 3 seconds.

No GPU Required

Uses cloud STT APIs (OpenAI Whisper, Deepgram). Run on any server without expensive hardware.

Multi-tenant

Each customer gets their own trained classifier, thresholds, and settings. Full isolation.

VICIdial Ready

Drop-in replacement for Asterisk AMD. EAGI integration with simple configuration.

Training Dashboard

Upload audio samples, label them, and retrain your model. Improve accuracy over time.

Analytics

Call stats, usage tracking, and detection metrics. Know how your system is performing.

Real-Time Dashboard

Monitor your AMD performance with live statistics, usage tracking, and recent call activity all in one place.

VoiceDetect Dashboard showing call statistics, monthly usage, and recent calls
350

Total Calls

Track all processed calls

82%

Machine Detected

High accuracy detection

881ms

Avg Processing

Sub-second latency

10K

Monthly Minutes

Usage quota tracking

VoiceDetect Call Logs with search, filter, and export functionality

Comprehensive Call Logs

Search, filter, and analyze every call processed by VoiceDetect.

Powerful Search & Filter

  • Filter by Called Number, Caller ID, Lead ID, or Unique ID
  • Filter by decision type (Human/Machine)
  • Date range filtering for historical analysis
  • Export results to CSV for reporting

Detailed Call Information

  • Full transcript of detected speech
  • Confidence percentage for each classification
  • Processing latency (typically 200-800ms)
  • VICIdial integration data (Lead ID, CID)
VoiceDetect Training interface with ML model management

Train Your Own Classifier

Upload samples, label them, and train a custom ML model for your specific use case.

Self-Service Training

  • Upload audio samples or text transcripts
  • Label samples as Human or Machine
  • One-click model training
  • Version control for trained models
53
Training Samples
100%
Model Accuracy
v1
Active Model
VoiceDetect Evaluate interface for testing classifier

Test Your Classifier

Evaluate your trained model with text input or voice uploads before going live.

Two Ways to Test

  • Text Input: Enter a transcript directly to see how it would be classified
  • Voice Upload: Upload an audio file to test real speech detection

Instant Feedback

  • See classification result (Human/Machine)
  • View confidence percentage
  • Debug edge cases before deployment
  • Fine-tune your training data based on results
VoiceDetect Provisioning with one-command installation

One-Command Installation

Deploy VoiceDetect to your ViciBox/VICIdial servers with a single command.

curl -sSL "https://..." | sudo bash

What Gets Installed

  • Python websocket-client library
  • Configuration file at /etc/voicedetect.conf
  • AGI script at /var/lib/asterisk/agi-bin/
  • Dialplan extension 8370 for AI AMD
  • Heartbeat monitoring (every minute)
  • voicedetect-test command for verification

Server Monitoring

  • Track all installed servers in one place
  • Online/offline status with heartbeat
  • View server details (IP, Asterisk version)
  • Last seen timestamp for troubleshooting
VoiceDetect Settings with AMD configuration options

Fine-Tune AMD Behavior

Customize timing, confidence thresholds, and silence handling to match your call center's needs.

Timing Settings (VAD-based)

  • Silence Timeout: Dead air bailout (default 2s)
  • Speech Processing: Time after speech starts (default 3s)
  • Max Total: Hard limit regardless of speech (default 6s)

Decision Thresholds

  • Min Confidence: Accept decision only above this threshold (default 60%)
  • Silence Handling: Treat no speech as Human or Machine

Expected Decision Times

  • No speech (dead air): ~2s
  • Immediate talker: ~3.5s
  • Delayed talker: up to 5s
  • Hard limit: 6s max

How It Works

1

Call Connects

When a call is answered, audio is streamed to VoiceDetect via WebSocket in real-time.

2

Speech Analysis

Voice Activity Detection (VAD) identifies speech. Audio is transcribed using cloud STT APIs.

3

ML Classification

Your tenant-specific classifier analyzes the transcript and determines human vs machine.

4

Instant Decision

Result returned via WebSocket. Route to agent or hangup - all within ~3 seconds.

Ready to Get Started?

VoiceDetect AMD is available as a hosted service or self-hosted solution. Contact us to discuss your requirements and get pricing.

Get in Touch