Audio Source Separation Using Residual Signal Re-separation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing audio source separation algorithms fail to effectively separate audio sources with overlapping components, as they assume independent sources, leading to suboptimal performance due to unmatched audio components in overlapping areas.
Innovation Solution
A method using Nonnegative Matrix Factorization-Expectation Maximization (NMF-EM) for initial separation, followed by a second separation operation on residual signals, where parameters are updated and weighted based on the mixed audio signal and residual signal power averages to improve source separation by accounting for overlapping areas.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If traditional audio source separation algorithms assume independent audio sources, then the separation process is simpler, but the separation performance deteriorates when audio sources have overlapping components
Solution Approach 1:
The separation process is divided into multiple stages: first separating audio sources assuming independence, then identifying residual signals containing overlapping components, and finally re-separating these residuals. This multi-stage segmentation approach allows the algorithm to handle overlapping components while maintaining manageable complexity at each stage.
Solution Approach 2:
The algorithm performs an initial separation that may be incomplete (partial action), identifies the remaining overlapping components as residuals, and then performs an additional separation pass on these residuals. This excessive action beyond a single separation pass ensures complete handling of both independent and overlapping components.
2Loss of time
If a single separation operation is performed on the mixed audio signal, then the processing time is shorter, but the separation completeness deteriorates due to unseparated residual signals
Solution Approach 1:
The separation process is performed periodically in multiple iterations: first separation to obtain initial sources and residuals, then second separation on the residuals to recover overlapping components. This periodic multi-pass approach ensures complete separation while keeping each individual pass computationally efficient.
Solution Approach 2:
The first separation operation is performed as a preliminary action to extract the majority of independent audio sources before addressing the more difficult overlapping components in the residual signals. This preliminary separation reduces the complexity of the subsequent residual separation.
Data Source
AI summary
A method for separating audio sources and an audio system using the same are provided. The method introduces the concept of a residual signal to separate a mixed audio signal into audio sources, and separates an audio signal corresponding to at least two of the audio sources as a residual signal and processes the audio signal separately. Therefore, audio separation performance can be improved. In addition, the method re-separates a separated residual signal and adds the separated residual signals to corresponding audio sources. Therefore, audio sources can be separated more safely.