Could you help to explain your heig2.m computation please?

[zcy618]

hi dear friend:
I checked your heig2.m file, but I could not understand what does this file do please?
I guess this file is doing "the generalized eigenvalue decomposition", but from your computation steps, I am not able to match the the generalized eigenvalue decomposition equation, and this file is only to support 2x2 matrix.
Thanks.

[nay0648]

Hi, thanks for watching. Heig2.m is used to solve the 2 x 2 generalized eigendecomposition problem for 2 x 2 Hermitian matrices. It only supports 2 x 2 Hermitian matrices, since that is enough for the two sources separation problems. Since the problem scale is small, eigenvalues and eigenvectors can be solved according to the definition of GEV problem, which yields the characteristic equation (1) in the attachment. Equation (1) means the determinant of (2) is zero, then, eigenvalues lambda can be solved from the quadratic equation in one variable. After this, eigenvalues are solved according to the equation (18) of: Ono, Nobutaka. "Fast stereo independent vector analysis and its implementation on mobile phone." IWAENC 2012; International Workshop on Acoustic Signal Enhancement. VDE, 2012. Best wishes! At 2021-01-20 07:46:43, "zcy618" <[email protected]> wrote: hi dear friend: I checked your heig2.m file, but I could not understand what does this file do please? I guess this file is doing "the generalized eigenvalue decomposition", but from your computation steps, I am not able to match the the generalized eigenvalue decomposition equation, and this file is only to support 2x2 matrix. Thanks. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or unsubscribe.

[zcy618]

Hi, thanks for watching. Heig2.m is used to solve the 2 x 2 generalized eigendecomposition problem for 2 x 2 Hermitian matrices. It only supports 2 x 2 Hermitian matrices, since that is enough for the two sources separation problems. Since the problem scale is small, eigenvalues and eigenvectors can be solved according to the definition of GEV problem, which yields the characteristic equation (1) in the attachment. Equation (1) means the determinant of (2) is zero, then, eigenvalues lambda can be solved from the quadratic equation in one variable. After this, eigenvalues are solved according to the equation (18) of: Ono, Nobutaka. "Fast stereo independent vector analysis and its implementation on mobile phone." IWAENC 2012; International Workshop on Acoustic Signal Enhancement. VDE, 2012. Best wishes! At 2021-01-20 07:46:43, "zcy618" [email protected] wrote: hi dear friend: I checked your heig2.m file, but I could not understand what does this file do please? I guess this file is doing "the generalized eigenvalue decomposition", but from your computation steps, I am not able to match the the generalized eigenvalue decomposition equation, and this file is only to support 2x2 matrix. Thanks. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or unsubscribe.

hi nay0648:
Thanks for your reply, I still have one question: the output has two channel data, but the required data is randomly in one of the two channel, for example, the input data has voice data and music, the output could seperate voice and music excellently, but the voice sometimes in channel 0, sometimes channel 1. So my question is that how could I get to know the required data in which channel, or how could I fix the voice in one channel?
Thanks

[nay0648]

Hi, original IVA cannot solve the channel order ambiguity. To solve this problem, some extra information should be involved. E.g. a, keyword spotting (KWS) system can be concatenated to the IVA output, the channel can be selected by the predefined keyword. In addition, approaches like Nesta, Francesco, and Zbyněk Koldovsk?. "Supervised independent vector analysis through pilot dependent components." Acoustics, Speech and Signal Processing (ICASSP), 2017 IEEE International Conference on. IEEE, 2017. Can be used. Some pilot information can be introduced, the pilot information is highly correlated to the target signal presence probability, which can be used to guide the IVA convergence to desired channel order. At 2021-02-09 08:51:56, "zcy618" <[email protected]> wrote: Hi, thanks for watching. Heig2.m is used to solve the 2 x 2 generalized eigendecomposition problem for 2 x 2 Hermitian matrices. It only supports 2 x 2 Hermitian matrices, since that is enough for the two sources separation problems. Since the problem scale is small, eigenvalues and eigenvectors can be solved according to the definition of GEV problem, which yields the characteristic equation (1) in the attachment. Equation (1) means the determinant of (2) is zero, then, eigenvalues lambda can be solved from the quadratic equation in one variable. After this, eigenvalues are solved according to the equation (18) of: Ono, Nobutaka. "Fast stereo independent vector analysis and its implementation on mobile phone." IWAENC 2012; International Workshop on Acoustic Signal Enhancement. VDE, 2012. Best wishes! At 2021-01-20 07:46:43, "zcy618" [email protected] wrote: hi dear friend: I checked your heig2.m file, but I could not understand what does this file do please? I guess this file is doing "the generalized eigenvalue decomposition", but from your computation steps, I am not able to match the the generalized eigenvalue decomposition equation, and this file is only to support 2x2 matrix. Thanks. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or unsubscribe. hi nay0648: Thanks for your reply, I still have one question: the output has two channel data, but the required data is randomly in one of the two channel, for example, the input data has voice data and music, the output could seperate voice and music excellently, but the voice sometimes in channel 0, sometimes channel 1. So my question is that how could I get to know the required data in which channel, or how could I fix the voice in one channel? Thanks — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe.