There is a first toy experiment with only 16 taps RIR, which is too few to model any kind of a real room situation. It unfortunately lacks sufficient experimental validation. The approach looks reasonable for the use case of a high amount of double-talk. The paper proposes the use of the Aux-ICA source separation method for AEC, and finds that it can be seen as a weighted RLS method. REVIEWER #3 Detailed Comments for Authors This could make positive impacts on the academic community.

To facilitate reproducible research, I suggest that the authors release the related source codes on, the website of the authors' research group, or similar website.

Low-complexity RLS algorithms using dichotomous coordinate descent iterations', IEEE Transactions on Signal Processing, vol. Also, the dichotomous coordinate descent approach for solving normal equations should be mentioned (e.g. 505-508 and Fast block exact Gauss-Seidel pseudo affine projection algorithm, IEE Electronics Letters, Oct. fast version from Combined echo and noise cancellation based on Gauss-Seidel pseudo affine projection algorithm,” Proc. The authors should mention the affine projection algorithms that link the NLMS and RLS algorithms (e.g. There are a few inconsistencies in the reference section that should be corrected.The numerical complexity of the proposed approach is not presented and it is not compared with the considered competing algorithms.Some assertions from the paper are not supported by references and the numerically efficient affine projection algorithms mentioned below are not introduced.why the step size of NLMS is 0.05 etc, filter length is 16). The choice of some parameters is not fully explained (e.g.The paper is well written having plenty of relevant figures and tables.

REVIEWER #2 Detailed Comments for Authors Quality of References - Reproducibility - = In Section 2.1, what is the meaning of N source images? Conclusion, “which both near-end and far-end signals coexists…” => “coexist”. Section 4.2, “which is accordance with…” => “which is accordant with…”. For example, Section 2.1, “The microphone signal x is consist of…” => “consists of…”. Explanation - Quality of the paperĮnglish should be revised carefully. The authors are suggested to explain the phenomenon. (9)ğrom Table 2, the performance of Aux is inferior to NLMS and RLS in terms of PESQ and Average Misalignment despite AuxP can substantially improve the performance. (8)ğigure 2 seems to be rather blurry and the curve of Aux before 10s can not be observed. Machine Intelligence Technology, Alibaba Group Yueyue Na, Ziteng Wang, Zhang Liu, Yun Li, Biao Tian, Qiang Fu A New Perspective of Auxiliary-Function-Based Independent Component Analysis in Acoustic Echo Cancellation

Acoustic echo cancellation meaning