Media Signature Generation via Frequency-Domain Curve Fitting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing media monitoring techniques face challenges in generating computationally efficient and robust signatures for identifying media signals, particularly in environments with noise and varying data capture methods, which are too intensive for mobile devices.
Innovation Solution
The method involves transforming audio signals into frequency-domain representations using a polyphase quadrature filter and applying a modified discrete cosine transform for energy compaction, followed by curve fitting to generate a signature function, which is then compressed into a set of angles representing the media block, enabling efficient and robust signature generation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If existing signature generation techniques are used, then media identification capability is achieved, but computational intensity becomes too high for mobile devices
Solution Approach 1:
The audio signal is divided into multiple blocks, and each block is processed independently to generate corresponding signature functions. This segmentation allows the complex signature generation task to be distributed across smaller, more manageable units, reducing the computational burden on mobile devices while maintaining identification accuracy.
Solution Approach 2:
The patent transforms audio signals from the time domain to the frequency domain using polyphase quadrature filters and modified discrete cosine transforms. This parameter transformation converts the signal representation into a format that enables energy compaction and more efficient signature function generation, significantly reducing computational requirements while preserving media identification capability.
2Reliability
If existing signature generation techniques are used, then media identification is possible, but robustness to noise and data capture variations deteriorates
Solution Approach 1:
The patent applies polyphase quadrature filters and modified discrete cosine transforms as preliminary processing steps before signature generation. These transformations pre-process the audio signal to compact energy and reduce the impact of noise and variations, creating a more robust foundation for subsequent signature function generation and media identification.
Solution Approach 2:
The patent generates multiple signature functions from different blocks of the audio signal, creating redundant representations that can compensate for noise and variations. By having multiple signature copies derived from overlapping or adjacent blocks, the system can select or combine the most accurate representations, improving robustness without sacrificing identification accuracy.
3Productivity
If existing signature generation techniques are used, then media monitoring is achieved, but data compression efficiency is insufficient
Solution Approach 1:
The patent applies modified discrete cosine transforms to convert audio blocks into the frequency domain, where energy can be compacted into fewer significant coefficients. This parameter transformation enables aggressive data compression by retaining only the most important frequency components, significantly reducing the data quantity needed for media monitoring while preserving identification capability.
Solution Approach 2:
The patent extracts and retains only the most significant energy components from the frequency-domain representation after transformation. By identifying and keeping only the essential coefficients that carry meaningful information about the audio signal, the system achieves high data compression efficiency while maintaining the productivity of media monitoring operations.
Data Source
AI summary
Example signature generation methods disclosed herein include transforming a block of signal samples from a time-domain representation to a frequency-domain representation including multiple frequency bands. Disclosed example signature generation methods also include fitting a signature function to at least a subset of the frequency bands. Disclosed example signature generation methods further include generating a set of signature values representative of the block of signal samples, the signature values generated based on a set of angles calculated using the signature function.