Ensemble-Based Audio Similarity Detection via Indirect Path Scoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting similarities between audio data items, such as cover songs, are computationally expensive and inefficient, struggling to identify versions that significantly differ from the original recording while still corresponding to the same underlying composition, making it difficult to enforce copyright laws and monitor user uploads effectively.
Innovation Solution
A computer-implemented method that calculates pairwise similarity scores between audio data items and updates them based on overall similarity scores through an indirect path involving intermediate items, using a hierarchical clustering algorithm and squashing functions to efficiently identify matches across vast libraries of tracks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If prior art techniques are used to detect whether two tracks share the same underlying composition, then similarity detection is achieved, but computational cost becomes excessively high and scalability is lost
Solution Approach 1:
The patent segments the similarity detection process into multiple stages: initial filtering using metadata, pairwise similarity scoring for candidate pairs, and ensemble-based refinement. This segmentation allows the system to quickly eliminate non-matching tracks before applying computationally intensive analysis, thereby maintaining accuracy while improving efficiency at scale.
Solution Approach 2:
The patent applies partial action by calculating pairwise similarity scores only for candidate pairs identified through initial filtering, rather than computing all possible pairs. The ensemble method then applies refinement only where needed, avoiding excessive computation on clearly non-matching tracks while maintaining detection accuracy for borderline cases.
2Adaptability or versatility
If prior art techniques are used for cover version detection, then some similarities are identified, but the ability to identify cover versions that significantly differ from the original is lost
Solution Approach 1:
The patent introduces an ensemble of multiple similarity scoring methods as intermediaries. Instead of relying on a single direct comparison, the system uses multiple intermediate scoring approaches (metadata-based filtering, acoustic feature comparison, lyrical analysis) that can bridge gaps between significantly different cover versions and their originals, thereby improving detection capability while maintaining precision through consensus scoring.
Solution Approach 2:
The patent changes parameters by applying different similarity metrics and weighting schemes across the ensemble of methods. By varying the parameters used for comparison (different acoustic features, metadata weights, lyrical similarity thresholds), the system can adapt to detect cover versions with significant variations while maintaining accurate similarity scoring through the aggregation of multiple parameter-based approaches.
3Reliability
If comprehensive pairwise comparison is performed on all audio data items, then all possible matches are identified, but the computational complexity becomes unmanageable for vast libraries
Solution Approach 1:
The patent performs preliminary actions by conducting metadata-based filtering and initial acoustic feature comparison before executing the full pairwise similarity analysis. This preliminary stage pre-identifies candidate pairs that are worth detailed comparison, ensuring match detection completeness for potential matches while reducing computational complexity by eliminating clearly non-matching pairs in advance.
Solution Approach 2:
The patent applies partial action by computing comprehensive pairwise comparisons only for candidate pairs identified through preliminary filtering, rather than performing exhaustive comparisons on all possible pairs in the library. This approach maintains reliability for identifying true matches while keeping device complexity manageable through selective application of computational resources.
Data Source
AI summary
The present disclosure relates to the comparison of data items, such as media data items or streams, e.g. audio files or streams, to determine a similarity between the data items. A device receives a plurality, or ensemble, of data items, and calculates the pairwise similarity scores between a number of unique pairs of data items within the ensemble. For any given pair of data items, the pairwise similarity score is compared with an overall similarity score for an “indirect path” connecting the pair of data items via one or more intermediate data items. If the indirect path indicates a closer similarity between the data items, the pairwise score is updated based on the overall similarity score of the indirect path. The updated similarity score generated by this comparison then provides an improved indication of the similarity of the data items compared to the prior art method.