Sound Signal Parameter Extraction Using Whitened Spectral Sequences
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current techniques for sound signal processing face challenges in using Line Spectrum Pair (LSP) parameters due to their multiple values, making direct application in sound classification and section estimation difficult, whereas the proposed parameter η, a shape parameter for generalized Gaussian distribution, offers easier threshold-based processing and efficient encoding and decoding.
Innovation Solution
A parameter determination device and method that transforms time-domain sound signals into the frequency domain, estimates spectral envelopes, and determines the parameter η by approximating the generalized Gaussian distribution of the whitened spectral sequence, facilitating efficient encoding and decoding for sound classification and estimation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If LSP parameters are used for sound signal processing, then comprehensive spectral information is obtained, but threshold-based processing becomes difficult due to multiple values
Solution Approach 1:
The patent extracts a single shape parameter η from the multiple LSP values by fitting a generalized Gaussian distribution to the spectral envelope. This extraction transforms the multi-value LSP representation into a single-value η parameter that retains essential spectral characteristics while enabling simple threshold-based classification.
Solution Approach 2:
The patent changes the parameter representation from multiple LSP values to a single shape parameter η. By transforming the parameter space and using distribution fitting, the system maintains spectral information accuracy while converting to a parameter form that is more suitable for threshold-based operations.
2Measurement precision
If LSP parameters are used, then detailed spectral characteristics are captured, but processing complexity increases
Solution Approach 1:
The patent extracts the essential spectral characteristic into a single shape parameter η through distribution fitting. This extraction process simplifies the data structure from multiple LSP values to one parameter while preserving the core spectral information needed for classification and section estimation.
Solution Approach 2:
The patent transforms the parameter set from multiple LSP values to a single shape parameter η. This parameter change reduces processing complexity by minimizing the number of parameters while maintaining the ability to capture detailed spectral characteristics through the distribution-based approach.
3Ease of operation
If a single value parameter η is used, then threshold-based processing becomes easier, but spectral information representation may be simplified
Solution Approach 1:
The patent introduces a generalized Gaussian distribution as an intermediary model between the spectral envelope and the shape parameter η. This distribution-based intermediary allows the single parameter η to effectively represent the spectral characteristics by capturing the distributional properties of the spectral envelope, thus maintaining representation accuracy while enabling simple threshold operations.
Solution Approach 2:
The patent changes from representing spectral information with multiple LSP values to representing it with a single shape parameter η derived from distribution fitting. This parameter transformation maintains spectral information representation accuracy by utilizing the statistical properties of the spectral envelope while achieving the simplicity of single-value threshold-based processing.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A parameter determination device comprises: a spectral envelope estimating portion 42 performing estimation of a spectral envelope using a parameter η0 specified in a predetermined method, regarding the η0-th power of absolute values of a frequency domain sample sequence corresponding to a time-series signal as a power spectrum on the assumption that the parameter η0 and a parameter η are positive numbers; a whitened spectral sequence generating portion 43 obtaining a whitened spectral sequence which is a sequence obtained by dividing the frequency domain sample sequence by the spectral envelope; and a parameter acquiring portion 44 determining such a parameter η that generalized Gaussian distribution with the parameter η as a shape parameter approximates a histogram of the whitened spectral sequence.