calculating mvdr using Rv

[hangtingchen]

Thank you for your code. I have a question about function mvdr. https://github.com/snsun/cgmm_mvdr/blob/cfc88aec4e26a6536eb4642566db9701ee9d6369/test.m#L37 According to your description of function mvdr, i think that Ry should be used instead of Rv. Refer to the Eq(9) in

Higuchi T , Ito N , Araki S , et al. Online MVDR Beamformer Based on Complex Gaussian Mixture Model With Spatial Prior for Noise Robust ASR[J]. IEEE/ACM Transactions on Audio, Speech, and Language Processing, 2017, 25(4):780-793.

[snsun]

Hi，

In theory, for MVDR, using Rv is equivalent to use Ry. As for my experience, Rv can get better results.

Best

hangtingchen notifications@github.com 于2019年2月19日周二 下午7:16写道：

Thank you for your code. I have a question about function mvdr.

https://github.com/snsun/cgmm_mvdr/blob/cfc88aec4e26a6536eb4642566db9701ee9d6369/test.m#L37 According to your description of function mvdr, i think that Ry should be used instead of Rv. Refer to the Eq(9) in

Higuchi T , Ito N , Araki S , et al. Online MVDR Beamformer Based on Complex Gaussian Mixture Model With Spatial Prior for Noise Robust ASR[J]. IEEE/ACM Transactions on Audio, Speech, and Language Processing, 2017, 25(4):780-793.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/snsun/cgmm_mvdr/issues/3, or mute the thread https://github.com/notifications/unsubscribe-auth/AMS4Kb5j54kY27npMXKs9TrhW0mySh00ks5vO9z9gaJpZM4bCvId .

[hangtingchen]

Thanks a lot for your reply. I just found that for test_wav/test1, it seemed using Rv suffered from larger distortion than using Ry. In addition, could you tell me the paper of the theory using Rv and Ry. Thanks for your patience.

Hi， In theory, for MVDR, using Rv is equivalent to use Ry. As for my experience, Rv can get better results. Best hangtingchen notifications@github.com 于2019年2月19日周二 下午7:16写道： … Thank you for your code. I have a question about function mvdr. https://github.com/snsun/cgmm_mvdr/blob/cfc88aec4e26a6536eb4642566db9701ee9d6369/test.m#L37 According to your description of function mvdr, i think that Ry should be used instead of Rv. Refer to the Eq(9) in Higuchi T , Ito N , Araki S , et al. Online MVDR Beamformer Based on Complex Gaussian Mixture Model With Spatial Prior for Noise Robust ASR[J]. IEEE/ACM Transactions on Audio, Speech, and Language Processing, 2017, 25(4):780-793. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#3>, or mute the thread https://github.com/notifications/unsubscribe-auth/AMS4Kb5j54kY27npMXKs9TrhW0mySh00ks5vO9z9gaJpZM4bCvId .

[snsun]

For MVDR, w = minw {w*R{yy}w} with the constraint w^H w = 1; where y = x + n; When we estimate R{yy}, expection is used as eq9. R{yy} = E[y y^H] = E[x x^H] + E[n * n^H] = R{xx} + R{nn};

If we want to minimize wR_{yy}w without any distortion for desired signal x, we only need to minimize wR_{nn}w.

I can not remember the paper or book I have read. In many papers, R_{nn} is commonly adopted to estimate MVDR coefficients.

Best Sining

hangtingchen notifications@github.com 于2019年2月19日周二 下午7:55写道：

Thanks a lot for your reply. I just found that for test_wav/test1, it seemed using Rv suffered from larger distortion than using Ry. In addition, could you tell me the paper of the theory using Rv and Ry. Thanks for your patience.

Hi， In theory, for MVDR, using Rv is equivalent to use Ry. As for my experience, Rv can get better results. Best hangtingchen notifications@github.com 于2019年2月19日周二 下午7:16写道： … <#m1630122912176713930> Thank you for your code. I have a question about function mvdr. https://github.com/snsun/cgmm_mvdr/blob/cfc88aec4e26a6536eb4642566db9701ee9d6369/test.m#L37 According to your description of function mvdr, i think that Ry should be used instead of Rv. Refer to the Eq(9) in Higuchi T , Ito N , Araki S , et al. Online MVDR Beamformer Based on Complex Gaussian Mixture Model With Spatial Prior for Noise Robust ASR[J]. IEEE/ACM Transactions on Audio, Speech, and Language Processing, 2017, 25(4):780-793. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#3 https://github.com/snsun/cgmm_mvdr/issues/3>, or mute the thread https://github.com/notifications/unsubscribe-auth/AMS4Kb5j54kY27npMXKs9TrhW0mySh00ks5vO9z9gaJpZM4bCvId .

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/snsun/cgmm_mvdr/issues/3#issuecomment-465101018, or mute the thread https://github.com/notifications/unsubscribe-auth/AMS4KdypYZza69lar7yPeTInzG3TM_tEks5vO-YigaJpZM4bCvId .

[hangtingchen]

Got it. Much appreciated.

For MVDR, w = minw {w*R{yy}w} with the constraint w^H w = 1; where y = x + n; When we estimate R{yy}, expection is used as eq9. R{yy} = E[y y^H] = E[x x^H] + E[n n^H] = R{xx} + R{nn}; If we want to minimize wR{yy}w without any distortion for desired signal x, we only need to minimize wR{nn}*w. I can not remember the paper or book I have read. In many papers, R_{nn} is commonly adopted to estimate MVDR coefficients. Best Sining hangtingchen notifications@github.com 于2019年2月19日周二 下午7:55写道： … Thanks a lot for your reply. I just found that for test_wav/test1, it seemed using Rv suffered from larger distortion than using Ry. In addition, could you tell me the paper of the theory using Rv and Ry. Thanks for your patience. Hi， In theory, for MVDR, using Rv is equivalent to use Ry. As for my experience, Rv can get better results. Best hangtingchen @.*** 于2019年2月19日周二 下午7:16写道： … <#m1630122912176713930> Thank you for your code. I have a question about function mvdr. https://github.com/snsun/cgmm_mvdr/blob/cfc88aec4e26a6536eb4642566db9701ee9d6369/test.m#L37 According to your description of function mvdr, i think that Ry should be used instead of Rv. Refer to the Eq(9) in Higuchi T , Ito N , Araki S , et al. Online MVDR Beamformer Based on Complex Gaussian Mixture Model With Spatial Prior for Noise Robust ASR[J]. IEEE/ACM Transactions on Audio, Speech, and Language Processing, 2017, 25(4):780-793. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#3 <#3>>, or mute the thread https://github.com/notifications/unsubscribe-auth/AMS4Kb5j54kY27npMXKs9TrhW0mySh00ks5vO9z9gaJpZM4bCvId . — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#3 (comment)>, or mute the thread https://github.com/notifications/unsubscribe-auth/AMS4KdypYZza69lar7yPeTInzG3TM_tEks5vO-YigaJpZM4bCvId .