Caller Identification Using Voice Characteristic Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing caller identification methods fail to accurately distinguish between calls from swindlers and non-swindlers, leading to potential oversight of swindler calls and false identification of non-swindlers as swindlers.
Innovation Solution
A caller identification apparatus and method that utilizes a white list for close relative voice characteristics and a black list for swindler voice characteristics, analyzing voice data to determine if it matches either list or includes characteristics of multiple persons, thereby classifying calls into detailed types to identify potential swindler calls with high accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If voice characteristics are compared only against a swindler database (black list), then swindler calls can be detected, but new swindlers or unknown callers are overlooked and misclassified as normal calls
Solution Approach 1:
The patent divides the caller identification process into multiple stages: first checking against a white list of trusted contacts, then analyzing voice characteristics against a black list of known swindlers, and finally performing multi-person voice separation analysis for calls that don't match either list. This segmented approach ensures that each type of call is evaluated with the most appropriate method, improving both reliability and precision.
Solution Approach 2:
The patent introduces a new dimension of analysis by detecting whether multiple persons are speaking in a single call. This additional analytical dimension allows the system to identify coordinated fraud schemes where multiple swindlers communicate together, thereby improving detection accuracy without increasing false positives.
2Device complexity
If any call not matching the swindler database is determined as normal, then processing is simple, but calls from new swindlers are overlooked
Solution Approach 1:
The patent performs preliminary voice separation analysis before final classification. By first determining whether multiple persons are present in the call, the system can then apply appropriate detection strategies: single-person calls are checked against voice databases, while multi-person calls trigger additional fraud suspicion flags. This preliminary action improves detection accuracy without significantly increasing overall system complexity.
3Reliability
If calls with voice characteristics not matching close relatives are determined as swindler calls, then fraud prevention is enhanced, but non-swindlers are falsely identified as swindlers
Solution Approach 1:
The patent implements a dynamic, multi-stage determination process rather than a static single-threshold classification. The system adapts its analysis depth based on initial matching results: calls matching white or black lists receive standard processing, while intermediate cases trigger additional multi-person voice separation analysis. This dynamic approach reduces false positives while maintaining fraud detection effectiveness.
4Productivity
If simple alternative determination methods are used, then processing is efficient, but swindler calls are overlooked or non-swindlers are misidentified
Solution Approach 1:
The patent applies partial analysis to most calls (checking against voice databases) and reserves the more computationally intensive multi-person voice separation analysis for specific cases where initial matching fails. This partial action approach maintains high processing efficiency for the majority of calls while applying enhanced analysis only where necessary to improve accuracy.
Data Source
AI summary
A caller identification apparatus, in order to make, with higher precision, a sharp distinction between a call from a fraud and a call from a person who is not a fraud, comprises a storage means and a voice characteristic analysis means. The storage means stores both a white list with which first voice characteristic information, which is the voice characteristic information of closely related persons of a call recipient, is registered and a black list with which second voice characteristic information, which is the voice characteristic information of frauds, is registered. The voice characteristic analysis means acquires the voice data of a call, obtains third voice characteristic information, which is the voice characteristic information of the voice data, and determines whether the third voice characteristic information matches the first voice characteristic information or the second voice characteristic information. If the third voice characteristic information matches neither the first voice characteristic information nor the second voice characteristic information, the voice characteristic analysis means determines whether the third voice characteristic information includes the voice characteristic information of any multiple persons. If the third voice characteristic information includes the voice characteristic information of any multiple persons, the voice characteristic analysis means obtains a first determination result that the call is probably a call from said frauds.