Spectral Smoothing via Subband Representative Values
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for speech signal processing require extensive calculation for non-linear transformation of all spectrum samples, leading to high computational costs and potentially suboptimal speech quality when only partial samples are processed.
Innovation Solution
A spectrum smoothing apparatus and method that performs time-frequency transformation, subband division, calculation of representative values using arithmetic and geometric means, non-linear transformation, and frequency domain smoothing, reducing the computational load while maintaining high speech quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If non-linear transformation processing is performed for all samples of a spectrum acquired from a speech signal, then speech quality is improved, but the amount of calculation processing becomes enormous
Solution Approach 1:
The frequency spectrum is divided into multiple subbands, and only representative values (e.g., maximum values) from each subband are selected for non-linear transformation processing. This segmentation approach maintains speech quality by preserving the most significant spectral characteristics while dramatically reducing the number of samples requiring intensive computation.
Solution Approach 2:
Representative values are extracted from each subband of the frequency spectrum before applying non-linear transformation. By selecting only the most significant spectral components (such as maximum values per subband) rather than processing all spectrum samples, the invention reduces computational load while maintaining the essential features needed for high-quality speech reconstruction.
2Productivity
If only part of samples of a spectrum are extracted to reduce the amount of calculation processing, then computational load is reduced, but sufficiently high speech quality cannot be always achieved
Solution Approach 1:
Different processing approaches are applied to different parts of the spectrum based on their importance. Representative values (maximum values) are extracted from each subband to capture the most significant spectral characteristics, while other less critical samples are discarded. This local quality approach ensures that the most important spectral information is preserved for high-quality speech reconstruction.
Solution Approach 2:
Instead of processing all spectrum samples or using a fixed partial subset, the invention applies non-linear transformation to specifically selected representative values from each subband. This partial action focuses computational resources on the most critical spectral components, achieving sufficient speech quality with reduced calculation processing.
3Device complexity
If representative values of subbands are used for non-linear transformation, then computational complexity is reduced, but spectral smoothing effectiveness may be compromised
Solution Approach 1:
The frequency spectrum is segmented into multiple subbands, and representative values are extracted from each subband for non-linear transformation. This segmentation maintains spectral smoothing effectiveness by preserving the most significant spectral characteristics in each frequency region while reducing computational complexity through selective processing.
Solution Approach 2:
The invention changes the parameter representation from individual spectrum samples to representative values (such as maximum values) of subbands. This parameter transformation reduces computational complexity while maintaining spectral smoothing effectiveness, as the representative values capture the essential spectral characteristics needed for quality speech reconstruction.
Data Source
AI summary
Disclosed is a spectral smoothing device with a structure whereby smoothing is performed after a nonlinear conversion has been performed for a spectrum calculated from an audio signal, and with which the amount of processing calculation is significantly reduced while maintaining excellent audio quality. With this spectral smoothing device, a sub band division unit (102) divides an input spectrum into multiple sub bands; a representative value calculation unit (103) calculates a representative value for each sub band using an arithmetic mean and a geometric mean; with respect to each representative value, a nonlinear conversion unit (104) performs a nonlinear conversion the characteristic of which is further emphasized as the value increases; and a smoothing unit (105) that smoothes the representative value which has undergone the nonlinear conversion for each sub band, at the frequency domain.