Answering Machine Detection Using Greeting Metadata Profiles
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Contact centers face challenges in accurately distinguishing between live person and automated voice responses, leading to inefficiencies and increased costs due to incorrect agent connections and call terminations.
Innovation Solution
Implementing a computer-controlled apparatus and method that uses a combination of timer-based and real-time speech analytics to analyze greetings, employing optimized AMD parameter sets and meta-data analysis to improve the accuracy of answering machine detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional AMD algorithms are used to detect answering machines, then some level of accuracy is achieved, but agent time is wasted due to incorrect dispositions
Solution Approach 1:
The system performs preliminary AMD analysis on a subset of calls to build and refine number-specific parameter sets before full deployment. This preliminary action creates optimized detection profiles that reduce false positives and improve accuracy on subsequent calls, thereby reducing agent time waste.
Solution Approach 2:
The system dynamically adjusts AMD parameters based on call outcomes and greeting patterns. By changing parameters such as detection thresholds and analysis windows based on accumulated data, the system improves accuracy over time while reducing incorrect dispositions that waste agent time.
2Productivity
If AMD accuracy is increased to reduce wasted agent time, then productivity improves, but the complexity of the detection system increases
Solution Approach 1:
The AMD system is segmented into modular components: real-time speech analytics engine, parameter set manager, number-specific profile database, and disposition recommendation system. This segmentation allows the system to achieve high accuracy through multiple specialized functions while maintaining manageable complexity through clear separation of concerns.
Solution Approach 2:
The system creates simplified copies of greeting patterns and stores them as number-specific parameter sets. Instead of analyzing every call in real-time with full complexity, the system uses compressed representations (parameter sets) that capture essential patterns, reducing computational complexity while maintaining accuracy.
3Measurement precision
If real-time speech analytics are implemented to improve AMD accuracy, then detection precision increases, but processing time and computational resources increase
Solution Approach 1:
The system performs partial speech analytics by analyzing only the most discriminative features of greetings rather than complete linguistic analysis. This partial action approach achieves sufficient precision for AMD while significantly reducing processing time and computational resource requirements.
Solution Approach 2:
Speech analytics models and parameter sets are pre-computed and stored during off-peak periods. This preliminary action allows the real-time system to quickly retrieve and apply pre-analyzed patterns rather than performing full analytics during call handling, maintaining precision while reducing processing time.
Data Source
AI summary
Answering machine detection (“AMD”) processes in a contact center are improved by obtaining and storing AMD meta-data about a known greeting from an automatic voice messaging capability (“AVMC”) on a telephone call to a known number. The AMD meta-data is used in subsequent calls to that known number and the greeting detected is analyzed using the AMD meta-data to make a comparison determination if AMD meta-data obtained from the current greeting matches that stored so as to determine whether the current greeting originated from an AVMC or from a live person. In certain embodiments, a real-time speech analytics (“RTSA”) system is used for processing the greeting to obtain the AMD meta-data which is stored and used when comparing subsequently obtained AMD meta-data. Calls to telephone numbers for which there is no stored AMD meta-data results in analyzing the greeting in order to obtain and store AMD meta-data.