, PSDs v s (n, f ), v z (n, f ) and v b (n, f ), we use the same procedure as in Section 4.1. Here, the residual echo z(n, f ) is the only latent signal
, Note that the denition of the 5 components of the early near-end signal s e (t) is an extension of
,
, Cascade : a cascade approach where the echo cancellation lter H(f ), the dereverberation lter G(f ) and the Wiener postlter W se (n, f ) are estimated and applied one after another. Echo cancellation relies on SpeexDSP 1 , which implements Valin's adaptive approach and is particularly suitable for time-varying conditions, Baselines Hereafter we denote our joint NN-supported approach as NN-joint. We compare it with four baselines: 1. Togami : our implementation of Togami et al.'s approach, vol.2
, NN-parallel : the variant of NN-joint where the echo cancellation lter H(f ) and the dereverberation lter G(f ) are applied in parallel as Togami et al.'s approach
, NN-cascade : the variant of Cascade where the echo cancellation lter H(f ) is estimated using the NN-supported approach similar to NN-joint (see Section 6) instead of Valin's adaptive approach. As WPE dereverberates similarly to its NN-supported counterpart in the multichannel case [16], NN-cascade corresponds to a cascade variant of NN-joint which estimates each lter separately using NN-supported optimization algorithms
Joint DNN-based multichannel reduction of echo, reverberation and noise, Speech, and Language Processing ,
Speech dereverberation based on variance-normalized delayed linear prediction, IEEE Transactions on Audio, Speech, and Language Processing, vol.18, issue.7, p.17171731, 2010. ,
Blind separation and dereverberation of speech mixtures by joint optimization, IEEE Transactions on Audio, Speech, and Language Processing, vol.19, issue.1, p.6984, 2011. ,
Multichannel audio source separation with deep neural networks, IEEE/ACM Transactions on Audio, Speech, and Language Processing, vol.24, issue.9, p.16521664, 2016. ,
URL : https://hal.archives-ouvertes.fr/hal-01163369
Multichannel nonnegative matrix factorization in convolutive mixtures for audio source separation, IEEE Transactions on Audio, Speech, and Language Processing, vol.18, issue.3, p.550563, 2010. ,
Generalization of multi-channel linear prediction methods for blind MIMO impulse response shortening, IEEE Transactions on Audio, Speech, and Language Processing, vol.20, issue.10, p.27072720, 2012. ,
Under-determined reverberant audio source separation using a full-rank spatial covariance model, IEEE Transactions on Audio, Speech, and Language Processing, vol.18, issue.7, p.18301840, 2010. ,
URL : https://hal.archives-ouvertes.fr/inria-00541865
Multichannel music separation with deep neural networks, Proc. EUSIPCO, p.17481752, 2016. ,
URL : https://hal.archives-ouvertes.fr/hal-01334614
Scalable audio separation with light kernel additive modelling, Proc. ICASSP, p.7680, 2015. ,
URL : https://hal.archives-ouvertes.fr/hal-01114890
Joint DNN-based multichannel reduction of echo, reverberation and noise: Supporting document, Inria, 2019. ,
On adjusting the learning rate in frequency domain echo cancellation with double-talk, IEEE Transactions on Audio, Speech, and Language Processing, vol.15, issue.3, p.10301034, 2007. ,
Multiple-input neural network-based residual echo suppression, Proc. ICASSP, p.231235, 2018. ,
URL : https://hal.archives-ouvertes.fr/hal-01723630
, Available: http, 2008.
SDR half-baked or well done, Proc. ICASSP, p.626630, 2019. ,
Simultaneous optimization of acoustic echo reduction, speech dereverberation, and noise reduction against mutual interference, IEEE/ACM Transactions on Audio, Speech, and Language Processing, vol.22, issue.11, p.16121623, 2014. ,
Neural network-based spectrum estimation for online WPE dereverberation, pp.384388-9303, 2017. ,