Anomalous Telephone Call Detection via Audio Embedding Vectors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting interconnect bypass fraud in telecommunication networks are slow, prone to false positives, and can be circumvented by bypass operators, necessitating a fast, accurate, and resource-efficient detection solution.
Innovation Solution
A computer-implemented method that analyzes call audio streams to detect anomalies, specifically interconnect bypass fraud, by deriving a set of features, converting them into an embedding vector, and using a back-end classifier to determine the probability of fraud, all while minimizing resource consumption.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional detection methods (test calls, call profiling, location detection) are used, then detection capability is provided, but detection speed is slow and false positives occur
Solution Approach 1:
The patent extracts and analyzes only the most critical audio features (MFCCs, spectral characteristics, temporal patterns) from the call audio stream, rather than processing the entire audio signal. This selective feature extraction enables fast detection while maintaining high accuracy in identifying interconnect bypass fraud patterns.
Solution Approach 2:
The system performs preliminary analysis of audio features during the call setup and initial phases, before fraud can fully manifest. By detecting anomalous patterns early in the call lifecycle, the system achieves fast detection without requiring complete call monitoring, thus reducing detection time while maintaining precision.
2Speed
If detection methods are made faster, then detection speed improves, but reliability and accuracy may deteriorate
Solution Approach 1:
The patent segments the audio stream into distinct temporal segments and analyzes different audio characteristics (spectral, temporal, statistical features) separately. This segmentation allows the system to process audio data in parallel and make rapid decisions about fraud detection while maintaining comprehensive analysis for reliability.
Solution Approach 2:
The system dynamically adjusts detection parameters and feature weighting based on call context and anomaly probability. By adapting analysis depth and computational resources based on initial findings, the system achieves fast detection for high-probability fraud cases while maintaining thorough analysis for borderline cases, thus balancing speed and reliability.
3Measurement precision
If comprehensive detection is performed, then detection completeness improves, but resource consumption increases
Solution Approach 1:
The patent applies partial action by analyzing only the necessary audio features required for fraud detection rather than processing all possible audio characteristics. This selective approach ensures complete detection of fraud patterns while minimizing computational resources by ignoring irrelevant audio data.
Solution Approach 2:
The system extracts and processes only the most discriminative audio features (MFCCs, spectral centroid, zero-crossing rate, temporal patterns) that are most indicative of interconnect bypass fraud. This feature selection enables comprehensive fraud detection with reduced computational overhead, as the system focuses resources only on the most informative audio characteristics.
Data Source
AI summary
Computer-implemented detection of anomalous telephone calls, for example detection of interconnect bypass fraud, is disclosed. A telephone call associated with user devices is analyzed remote from the user devices. A first set of multiple features, for example Mel Frequency Cepstral Coefficients, is derived from a call audio stream. The first set is converted to an embedding vector, for example via a model based on a Universal Background Model comprising a Gaussian Mixture Model, which model is preferably configured based on a training plurality of first sets of multiple features derived form a corresponding training plurality of audio streams. Occurrence, or probability of occurrence, of an anomalous telephone call is determined based on the embedding vector, for example via a back-end classifier, such as a Gaussian Backend Model, which classifier is preferably configured based on labels associated with the training plurality of audio streams.