Answering Machine Detection Using Call Pickup Timing
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 utilizes timer-based and linguistic analysis of greetings, combined with real-time speech analytics, to determine whether a call has been answered by a live person or an automated voice messaging system, optimizing AMD parameters for specific scenarios and telephone numbers.
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 still wasted due to incorrect detections
Solution Approach 1:
The system dynamically adjusts AMD parameters based on call characteristics and historical data. The AMD threshold and sensitivity are not fixed but adapt in real-time based on the specific call context, allowing the system to optimize between false positives and false negatives dynamically during call operations
Solution Approach 2:
The system implements feedback loops where agent corrections and call outcomes are used to continuously refine AMD algorithms. When agents correct AMD decisions or when call outcomes indicate detection errors, this feedback is fed back into the system to improve future detection accuracy and reduce time waste
2Productivity
If AMD accuracy is increased to reduce agent time wastage, then productivity improves, but system complexity increases
Solution Approach 1:
The AMD system is segmented into multiple independent analysis components: acoustic feature extraction, linguistic analysis, timing pattern recognition, and machine learning classification. Each segment handles a specific aspect of detection, allowing the system to achieve high accuracy through modular, manageable components rather than a monolithic complex system
Solution Approach 2:
The system introduces intermediary layers between the raw audio input and the final detection decision, including feature extraction intermediaries and multiple analysis stages. These intermediaries process and transform data in manageable steps, reducing the complexity of the overall detection system while maintaining high productivity
3Measurement precision
If multiple AMD parameters are analyzed to improve detection accuracy, then distinction between live person and machine improves, but processing time increases
Solution Approach 1:
The system employs periodic analysis stages where different parameter sets are evaluated at specific intervals during the call. Instead of continuously analyzing all parameters simultaneously, the system periodically switches between different analysis modes (acoustic, linguistic, timing) to maintain accuracy while controlling processing time through structured temporal sampling
Data Source
AI summary
Answering machine detection (“AMD”) processes in a contact center are improved by obtaining and storing call pickup times regarding answered calls. The call pickup time is based on the time between detection of a signaling message indicating the call was offered to the remote interface and a signaling message indicating the call was answered. The value of the call pickup time may be useful to determine if an automatic voice messaging capability (“AVMC”) or a live human answered the call. In other embodiments, the call pickup time is used to generate a call pickup time weighting factor that is used to supplement the analysis of the initial audio greeting after the call has been answered to determine whether an AVMC or live person answered the call. The analysis can be used to determine whether the AVMC is an answering machine or a voice mail service.