Harmonic-Percussive Sound Separation Using Structure Tensor Orientation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing harmonic-percussive sound separation methods fail to accurately capture frequency modulated sounds, which carry tonal information, due to their reliance on strict horizontal and vertical structure assumptions in spectrograms, leading to leakage into the residual component.
Innovation Solution
The use of a structure tensor to determine predominant orientation angles in the magnitude spectrogram, allowing for the distinction between harmonic, percussive, and residual signal components, even in frequency modulated signals, by calculating local frequency changes and anisotropy measures.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If strict horizontal and vertical structure assumptions are used in spectrogram-based separation methods, then computational complexity is reduced and separation speed is improved, but frequency modulated sounds with tonal information leak into the residual component
Solution Approach 1:
The patent changes the classification parameters from strict horizontal/vertical orientation to orientation-aware classification using structure tensor angles. By calculating the predominant orientation angle at each spectrogram bin and comparing it against threshold ranges, the method adapts the classification criteria to accommodate frequency modulated sounds while maintaining computational efficiency.
Solution Approach 2:
The patent introduces dynamic orientation detection by computing structure tensor angles that adapt to local spectral patterns. Instead of fixed horizontal/vertical classification, the method dynamically determines the orientation of spectral structures at each time-frequency bin, allowing flexible classification that responds to the actual signal characteristics including frequency modulation.
2Measurement precision
If structure tensor calculation is used to determine predominant orientation angles, then accuracy of capturing frequency modulated sounds is improved, but computational complexity increases
Solution Approach 1:
The patent segments the spectrogram into discrete time-frequency bins and processes each bin independently using structure tensor calculation. This segmentation allows the complex orientation detection to be performed in a localized and systematic manner, reducing the overall computational burden compared to global processing approaches.
Solution Approach 2:
The patent applies local quality analysis by computing structure tensor angles specifically at spectrogram bins where tonal information is present. The orientation-aware classification is applied selectively based on local spectral characteristics, focusing computational resources on regions that require detailed analysis while simplifying processing in other regions.
3Measurement precision
If orientation-aware classification is used to distinguish harmonic components, then separation accuracy is improved, but device complexity increases
Solution Approach 1:
The patent replaces complex iterative optimization mechanisms with a direct orientation-based classification approach. Instead of using iterative algorithms to separate harmonic and percussive components, the method substitutes a straightforward classification rule based on structure tensor angle thresholds, significantly reducing algorithmic complexity while maintaining separation accuracy.
Data Source
AI summary
An apparatus for analysing a magnitude spectrogram of an audio signal is provided. The apparatus includes a frequency change determiner being configured to determine a change of a frequency for each time-frequency bin of a plurality of time-frequency bins of the magnitude spectrogram of the audio signal depending on the magnitude spectrogram of the audio signal. Moreover, the apparatus includes a classifier being configured to assign each time-frequency bin of the plurality of time-frequency bins to a signal component group of two or more signal component groups depending on the change of the frequency determined for the time-frequency bin.