Answering Machine Detection via Configurable Acoustic Parameters
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing answering machine detection (AMD) modules in voice solutions service providers face challenges in accurately and efficiently determining whether an outbound call is answered by a human or an answering machine, leading to incorrect call handling and a poor caller experience.
Innovation Solution
The implementation of an improved AMD module that processes call responses based on configurable parameters such as maximum time out, DTMF tone detection, background noise, silence duration, beep timeout, word count, and no speech detection, allowing for precise determination of whether the call recipient is human or a machine, and enabling customizable configurations through a web-enabled interface.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional AMD modules are used to determine whether a call recipient is human or machine, then the system can provide automated call handling, but the detection accuracy is insufficient leading to incorrect determinations
Solution Approach 1:
The AMD module is divided into multiple independent detection components that analyze different acoustic features separately: background noise detection, silence duration detection, DTMF tone detection, and speech content analysis. Each component processes specific parameters and contributes to the overall determination, allowing for more accurate and reliable classification of human versus machine callers.
2Measurement precision
If multiple detection parameters are used to improve detection accuracy, then the precision of determining call recipient type improves, but the processing time and system complexity increase
Solution Approach 1:
The system performs preliminary detection using quick-to-analyze parameters such as background noise levels, silence duration, and presence of DTMF tones during the initial phase of call handling. These preliminary results provide early indicators that can quickly identify obvious cases (e.g., answering machines with characteristic beeps or silence patterns), reducing the need for more time-consuming speech analysis in clear-cut scenarios.
Solution Approach 2:
The AMD module applies a tiered analysis approach where not all detection parameters are processed with equal depth for every call. For calls that show strong indicators of machine responses in preliminary detection, the system makes determinations based on partial analysis. For ambiguous cases, additional parameters are analyzed in greater detail to ensure accurate classification, thus optimizing the balance between processing time and detection precision.
3Adaptability or versatility
If the AMD module uses multiple configurable parameters for detection, then the system can be adapted to different client needs, but the device complexity increases
Solution Approach 1:
The AMD module is designed as a universal system that incorporates multiple detection capabilities (noise detection, silence detection, DTMF detection, speech analysis) within a single integrated architecture. This multi-functional design allows the same module to serve diverse client needs by enabling or disabling specific parameters through configuration, rather than requiring separate modules for different detection types. The configurable parameters include maximum time out, touch tone detection, noise thresholds, silence duration, beep timeout, word count, and no speech detection.
4Measurement precision
If the system processes call responses with multiple parameters and configurations, then the accuracy of call recipient determination improves, but the computational resources and processing overhead increase
Solution Approach 1:
The AMD module processes call responses in periodic time frames rather than continuously analyzing every audio sample. It divides the call response into discrete time windows and applies parameter detection at these periodic intervals, analyzing background noise, silence duration, and speech content at specific sampling points. This periodic processing reduces computational overhead while maintaining detection accuracy by capturing essential features at critical moments in the call interaction.
Data Source
AI summary
A call is placed to a communications device. A response to the call is received from a call recipient associated with the communications device. The received response is processed based on a set of parameters. Based on processing the received response, a determination is made whether the call recipient is a human or an answering machine. If the call recipient is determined to be a human, the call is handled in a first manner. If the call recipient is determined to be an answering machine, the call is handled in a second manner that is different from the first manner.