Introduction
This article provides an overview of the AMD (Answering Machine Detection) settings in Nixxis Contact Suite. It explores the technical parameters and customisation options available for optimising AMD settings based on specific requirements.
Depending on your telecom operator, the default detection settings may react differently. If this is the case, it is advised to duplicate the default settings and then adjust the parameters on the new settings.
Parameters
Below are the detection settings that can be adjusted from the administration to try to improve the results :
- Adapting the maximum detection time (default 5000 ms): this can be done via SQL, in the Admin database, in the AmdSettings table.
This is the time allowed for the system to decide if a call is being answered by a human or an answering machine.
Increasing this time can improve accuracy but may also increase wait time for callers.
- Initial silence duration
The length of silence at the beginning of a call before the system decides it's an answering machine.
Shortening this time can reduce false positives but might increase false negatives.
- Maximum number of words
This parameter helps in differentiating between a human greeting and an answering machine message based on the number of words spoken.
Adjusting this can help in identifying longer messages typical of answering machines.
- Rhythm Analysis
This sophisticated feature analyzes the voice's rhythm to differentiate between humans and machines.
However, it's also where the "uncertainty area" comes into play due to the variability in human speech patterns.
It is important to consider that:
- Regardless of the settings, there will always be a level of uncertainty inherent in how answering machine detection operates, particularly in analysing voice patterns
- Changes to the settings can also result in false detections of answering machines, which can be difficult to identify
- Additionally, the longer the allotted time for detection, the higher the quality of detection, but this can also lead to a delay perceived by the customer before being connected to an agent
You also have the option to select actions based on the following cases:
- If a person is detected
- If an answering machine is detected
- If the detection result is uncertain
For these cases, you can choose to:
- Play a recording
- Hang up the call
To configure the action of playing a message, you need to:
- Check the "Play recording" box in the appropriate case (in this case, when an answering machine is detected)
- Select or import the "Answering machine detection message" at the outbound activity level
If no action is specified (unchecked ), the call will be routed to the agent.
The Trade-Offs
Adjusting these settings involves a trade-off between detecting as many answering machines as possible (specificity) and ensuring live calls are not incorrectly flagged (sensitivity).
The key is to find a balance that suits your customer's operational needs and tolerance for risk (i.e., missing a live call vs. wasting time on an answering machine).
Example Configuration and Expected Results
Here’s a simplified adjustment strategy in your scenario where it has been experienced high false positives, with the AMD system frequently mistaking live calls for answering machines.
Before Adjustment
- Maximum Detection Time : 5000 ms
- Initial Silence Duration : 2500 ms
- Maximum Number of Words : 3
After Adjustment
- Reduce Maximum Detection Time to 4000 ms : A shorter window can force a quicker decision, potentially reducing the chance of false positives but may slightly increase false negatives
- Decrease Initial Silence Duration to 2000 ms : This can make the system less likely to interpret a brief pause by a human as an answering machine
- Increase Maximum Number of Words to 5 : This allows for longer human greetings to be recognized as such, reducing the likelihood of flagging them as answering machines
Expected Results
Reduced False Positives : The system becomes less likely to mistake live calls