Answering Machine Detection

Answering Machine Detection (AMD) analyzes the audio at the start of a call to determine whether a human or a machine answered. This allows you to handle voicemail systems differently from live answers.

What Is AMD

When AMD is enabled, the Cloudonix platform analyzes the audio stream at the beginning of each call. It looks for patterns that indicate:

• Human Answer: A live person answered the phone
• Machine Answer: An answering machine or voicemail system answered

Based on the detection result, you can configure different call handling behaviors.

Enabling AMD

Enable AMD on the Advanced tab of the campaign creation or edit form:

1. Toggle AMD to the ON position
2. Select the Mode (Enable or DetectMessageEnd)
3. Configure the detection parameters
4. Save the campaign

note

AMD adds a small delay before the call is connected to the routing destination. The detection timeout determines the maximum delay.

AMD Modes

The system supports two AMD modes:

Mode	Cloudonix Value	Behavior
Enable	Enable	Detects human vs machine. Proceeds with routing for humans. Can hang up on machines or route them differently.
Detect Message End	DetectMessageEnd	Waits for the end of the answering machine greeting before connecting the call. Useful when you want to leave a message.

Enable Mode

Use Enable mode when you want to:

• Connect live answers to your AI assistant immediately
• Hang up on machines without leaving a message
• Track answer types in your campaign statistics

DetectMessageEnd Mode

Use DetectMessageEnd mode when you want to:

• Leave a message on answering machines
• Ensure your AI assistant starts speaking after the beep
• Maximize message delivery rates

AMD Configuration Parameters

Fine-tune AMD behavior with these parameters:

Parameter	Default	Unit	Description
Detection Timeout	30	seconds	Maximum time to wait for AMD result before proceeding
Speech Threshold	1500	milliseconds	Minimum duration of speech to classify as human voice
Speech End Threshold	2500	milliseconds	Duration of silence after speech to determine the end of a greeting
Silence Timeout	3500	milliseconds	Initial silence period before any speech is detected

Parameter Details

Detection Timeout

• The maximum time AMD will analyze the call before making a decision
• If timeout is reached without a clear determination, the call proceeds as if human answered
• Longer timeouts increase accuracy but add delay

Speech Threshold

• Minimum amount of continuous speech required to trigger human detection
• Shorter thresholds may misclassify short machine greetings as human
• Longer thresholds may miss brief human greetings

Speech End Threshold

• Amount of silence after speech that indicates the end of a greeting
• Critical for DetectMessageEnd mode to know when the beep occurs
• Shorter thresholds may trigger too early; longer thresholds add delay

Silence Timeout

• Initial silence period that must pass before speech analysis begins
• Helps filter out connection delays and initial line noise
• Should be shorter than typical human greeting response time

How AMD Affects Call Flow

When AMD is enabled, the call flow changes:

1. Call initiated by Go Worker
2. Cloudonix dials the destination
3. Callee answers
4. AMD analysis begins (Detection Timeout window starts)
5. Audio patterns analyzed:
   - Live human: Short greeting, immediate response
   - Machine: Longer greeting, pause, beep
6. Result determined:
   - Human: Connect to routing destination immediately
   - Machine (Enable mode): Hang up or route differently
   - Machine (DetectMessageEnd mode): Wait for beep, then connect
7. Call proceeds based on result

AMD in Campaign Statistics

AMD results are tracked in your campaign data:

• Human Answers: Counted as successful connections
• Machine Answers: Counted separately for analysis
• Unknown/Timeout: Calls where AMD could not determine the answer type

Use these statistics to optimize your AMD parameters over time.

Best Practices

1. Test Different Parameters: AMD accuracy varies by region and phone system. Test with your target audience.

2. Consider Your Use Case:

   • Sales campaigns: Use Enable mode to focus on live answers
   • Notification campaigns: Use DetectMessageEnd mode to leave messages

3. Monitor Accuracy: Track human vs machine detection rates and adjust parameters if needed

4. Respect Voicemail: If using DetectMessageEnd mode, ensure your message is concise and starts immediately after connection

5. Balance Delay vs Accuracy: Longer detection timeouts improve accuracy but add delay to live answers

Troubleshooting

Symptom	Cause	Solution
Many false human detections	Speech threshold too low	Increase speech threshold
Many false machine detections	Speech threshold too high	Decrease speech threshold
Messages cut off	Speech end threshold too short	Increase speech end threshold
Long delays before connection	Detection timeout too long	Reduce detection timeout
AMD not triggering	AMD disabled or misconfigured	Verify AMD is enabled and parameters are set
Inconsistent results	Regional variations	Test with local phone systems and adjust parameters

AMD and Call Recording

When both AMD and call recording are enabled:

• Recording starts when the call is answered
• AMD analysis occurs during the initial part of the recording
• The recording captures the full call including any AMD-related delays

tip

Review recorded calls to evaluate AMD accuracy. Listen to how the system classified different answer types and adjust parameters accordingly.