Audio Content Identification Using Human Review Thresholds
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing content identification systems face challenges in accurately matching altered or rotated audio/video content due to vulnerabilities in fingerprinting algorithms, which can be exploited by hackers, leading to false negatives and inefficiencies in database searches.
Innovation Solution
Incorporating human judgment through crowdsourcing platforms like Amazon Mechanical Turk to assist in content identification by presenting unknown content to human reviewers for comparison with reference content, enhancing the accuracy of matches and overcoming algorithmic limitations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If fingerprinting algorithms are used for content identification, then automated matching speed is improved, but accuracy deteriorates due to vulnerabilities to alteration and rotation exploits
Solution Approach 1:
The patent introduces human reviewers as an intermediary between automated fingerprinting algorithms and final content identification results. When algorithmic confidence falls below a threshold, human reviewers examine the content to provide accurate identification, thereby resolving the contradiction between automated speed and accuracy by using humans to correct algorithmic errors caused by exploitation
Solution Approach 2:
The system implements feedback loops where identification results (including human reviews) are used to update and refine the fingerprinting algorithms. This continuous feedback improves algorithmic accuracy over time while maintaining automated processing speed, gradually reducing the need for human intervention
2Reliability
If human reviewers are used to improve identification accuracy, then false negatives are reduced, but processing time increases
Solution Approach 1:
The system applies human review partially - only to cases where algorithmic confidence falls below a predetermined threshold. Most straightforward cases are handled automatically without human intervention, while only ambiguous or potentially exploited cases receive human review, thus improving accuracy without proportionally increasing processing time for all cases
Solution Approach 2:
The system dynamically adjusts the confidence threshold parameter that triggers human review. By optimizing this parameter, the system balances the trade-off between using more human reviewers (improving reliability) and keeping processing time low, adapting the threshold based on performance metrics and resource availability
Data Source
AI summary
The disclosed technology generally relates to methods for identifying audio and video entertainment content. One claim recites network server comprising: an input for receiving data representing audio uploaded to said network server; memory for storing the data representing audio; one or more processors configured for processing the data representing audio to yield fingerprint data; memory for storing fingerprint data; one or more processors configured for: determining whether the fingerprint data incurs a potential match with the stored fingerprint data, the potential match indicating an unreliability in the match below a predetermined threshold; and issuing a call, upon a condition of unreliability in the match, requesting at least a first reviewer and a second reviewer to review the data representing audio; an interface for receiving results from the first reviewer and results from the second reviewer; and one or more processors configured for weighting results from the first reviewer differently than results from the second reviewer, and determining whether to allow public access to the data representing audio based at least in part on weighted results. Of course other combinations and claims are provided.