Content Fingerprints Using Averaged Spectrograms for Offset-Resilient Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing content matching techniques are undermined by signal offsets, leading to inaccurate fingerprint generation and reduced matching rates, especially when signals are shifted by a fraction of a frame, resulting in fingerprint collisions and mismatches.
Innovation Solution
Generate fingerprints using an averaged spectrogram and quantize peak positions to improve robustness, employing a bitOR function with selective bit shifting and quantization to enhance matching accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional fingerprint generation is used without averaging spectrograms, then the process is simpler and faster, but matching accuracy deteriorates when signals have offsets or are shifted by fractions of a frame
Solution Approach 1:
The patent applies preliminary action by generating multiple spectrograms at different time offsets (e.g., 0, 0.5, 1.0 frames ahead) before fingerprint generation, then averaging them to create a more robust reference fingerprint that is less sensitive to timing variations in the input signal
Solution Approach 2:
The patent changes the parameter of time offset by generating spectrograms at multiple different time offsets relative to the input signal, then combining these variations through averaging to produce a fingerprint that is invariant to small timing shifts
2Reliability
If signal sampling is performed with fixed frame alignment, then processing is more efficient, but matching reliability deteriorates when signals have sampling offsets
Solution Approach 1:
The system performs preliminary action by pre-calculating and storing multiple spectrograms at different time offsets for reference signals, so that when an input signal arrives, the appropriate pre-computed spectrograms can be quickly retrieved and averaged without real-time computation delays
Solution Approach 2:
The patent applies dynamics by making the fingerprint generation process adaptive to timing variations - instead of using a fixed time alignment, the system dynamically selects and averages multiple spectrograms at different offsets based on the characteristics of the input signal
3Measurement precision
If fingerprint generation uses exact peak positions, then fingerprint precision is higher, but fingerprint collisions increase when signals have minor variations
Solution Approach 1:
The patent changes the parameter of time offset and applies this transformation to the reference signal to generate multiple variations, then averages these variations to create a more robust fingerprint that is less sensitive to exact peak position variations in the input signal
Data Source
AI summary
Devices, systems, and methods are provided for using fingerprints of signals to match content. A method may include receiving, by a device, a first signal; generating a first value indicative of a first peak of a spectrogram for the first signal; generating a second value indicative of a second peak of the spectrogram; generating a third value indicative of a difference between a first time of the first peak and a second time of the second peak; generating a first quantized value based on the first value; generating a second quantized based on the second value; generating a first fingerprint based on the first quantized value, the second quantized value, and the third value; determining, based on a comparison of the first fingerprint to a second fingerprint, a match between the first fingerprint and the second fingerprint; and sending an indication of the match.