Sound Signal Detection Using Statistical Feature Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current signal detection methods fail to accurately search for similar sound or image signals in databases due to characteristic distortions and noise, leading to reduced accuracy and increased search time.
Innovation Solution
A system that calculates and compares feature vectors from target and stored signals, normalizes and quantizes these features to reduce noise influence, and selects statistically strong elements for comparison, using multidimensional vectors and scalar quantization to enhance search accuracy and efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional signal detection methods are used to search for similar sound or image signals in databases, then the search operation can be performed, but the accuracy is reduced greatly due to characteristic distortions and noise
Solution Approach 1:
The patent extracts only the statistically strong elements from the feature vectors by comparing each element's absolute value against the mean of absolute values. This extraction process removes weak or noisy features while retaining only the significant characteristic elements, thereby improving search accuracy by eliminating noise and distortion influence from the comparison process
Solution Approach 2:
The patent changes the parameter representation by transforming raw feature vectors into a new representation where each element is the ratio of the absolute value to the mean of absolute values. This parameter transformation normalizes the features and highlights statistically significant elements, enabling accurate signal detection even in the presence of characteristic distortions and noise
2Measurement precision
If multiple noises or distortions are considered by providing fluctuation appending step, then signal detection accuracy improves, but the amount of information increases requiring multiple target features
Solution Approach 1:
The patent segments the feature vector into individual elements and evaluates each element's statistical significance separately. By processing features individually and selecting only those exceeding the threshold, the system maintains detection accuracy for multiple noise types without requiring multiple separate target features, thus controlling the amount of information required
Solution Approach 2:
The patent transforms the feature representation by calculating the ratio of each element's absolute value to the mean of absolute values. This parameter change creates a normalized representation that inherently accounts for various noises and distortions through statistical evaluation, achieving accurate signal detection with a single target feature set rather than requiring multiple features for different noise types
3Measurement precision
If feature vectors with all elements are compared, then comprehensive comparison is performed, but the search time increases
Solution Approach 1:
The patent extracts only the statistically strong elements from the feature vectors by comparing each element against the mean of absolute values and selecting those that exceed the threshold. This extraction reduces the number of elements to be compared while maintaining comprehensive comparison of significant features, thereby reducing search time without sacrificing comparison quality
Solution Approach 2:
The patent applies partial action by comparing only the statistically significant elements rather than all elements in the feature vectors. By performing comparison on a subset of important features (those exceeding the threshold), the system achieves sufficient comprehensiveness for accurate detection while significantly reducing the computational burden and search time
Data Source
AI summary
A sound signal detection system of the present invention that searches for a portion of stored sound signals similar to a target sound signal, includes a stored feature calculation portion that calculates a stored feature from time-series data of the stored sound signals; a target feature calculation portion that calculates a target feature from time-series data of the target sound signal; a stored feature area selection portion that selects elements corresponding to statistics larger than a threshold from stored features and calculates stored area selection features generated from the selected elements; a target feature area selection portion that selects elements corresponding to statistics larger than a threshold from a target feature and calculates a target area selection feature generated from the selected element; and a feature comparison portion that sets a comparison segment in the stored area selection features and calculates a degree of similarity between comparison segments of both the target and stored area selection features.