Relative Transfer Function Estimation via Signal Sparsity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In high reverberation environments, the adaptability of single source models is compromised due to overlapping speaker spectra, making it difficult to estimate transfer functions effectively.
Innovation Solution
A device and method that compute a correlation matrix from multi-channel microphone signals, decompose it into signal space basis vectors, and use these to estimate relative transfer functions (RTFs) even in scenarios with overlapping speaker spectra by transforming the signals to make them sparse in the time direction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If single source model is used for RTF estimation, then estimation process is simple, but adaptability deteriorates in high reverberation environments where speaker spectra overlap
Solution Approach 1:
The patent segments the mixed speaker spectra into individual source components by computing M vectors from the correlation matrix eigenvectors, where each vector corresponds to a different sound source. This segmentation allows the system to handle multiple overlapping speakers separately, resolving the contradiction between simple processing and adaptability to complex acoustic environments.
Solution Approach 2:
The patent transforms the problem from spectral domain analysis to time-direction sparsity by determining coefficients that make the signal sparse in the time direction. This dimensional transformation enables effective separation of overlapping speakers without requiring complex spectral analysis, maintaining computational simplicity while improving adaptability.
2Ease of operation
If conventional RTF estimation methods are used, then processing is straightforward, but measurement precision deteriorates when multiple speakers are present
Solution Approach 1:
The patent performs preliminary decomposition of the correlation matrix into M eigenvectors before RTF estimation. This preliminary action creates a structured basis that facilitates accurate RTF estimation for multiple speakers, maintaining operational straightforwardness while improving measurement precision through systematic signal separation.
Solution Approach 2:
The patent changes the parameter representation by expressing RTFs in terms of eigenvector coefficients rather than direct spectral ratios. This parameter transformation enables accurate estimation of multiple overlapping speakers while keeping the processing framework similar to conventional methods, thus maintaining ease of operation while improving precision.
Data Source
AI summary
The transfer function estimation device includes: a correlation matrix computing unit 43 computing a correlation matrix of N frequency domain signals y(f,l); a signal space basis vector computing unit 44 obtaining M vectors v1(f), . . . , vM(f) from eigenvectors of the correlation matrix from highest in the order of corresponding eigenvalues; and a plural RTF estimation unit 45 determining ti(f), . . . , tM(f) that satisfy the relationship of Expression (1), determining a matrix D(f) that is not a zero matrix and that makes ui(f), . . . , uM(f) defined by Expression (2) sparse in a time direction, determining ci,1(f), . . . , cM,N(f) that satisfy the relationship of Expression (3), and outputting c1(f)/c1,j(f), . . . , cM(f)/cM,j(f) as a relative transfer function, where j is an integer of 1 or more and not more than N.