Final results
The below table gives the accepted papers in order of performance rank. Note that the paper acceptance is not well correlated to performance. Paper acceptance was decided with the normal INTERSPEECH peer-review process.
Place |
Performance Rank |
Team |
Authors |
Title |
1 |
6 |
Chinese Academy of Sciences |
Renhua Peng, Linjuan Cheng, Chengshi Zheng and Xiaodong Li |
|
2 |
8 |
Northwestern Polytechnical University |
Shimin Zhang, Yuxiang Kong, Shubo lv, Yanxin Hu and Lei Xie |
|
3 |
10 |
Evolve Technologies |
Lukas Pfeifenberger, Matthias Zöhrer and Franz Pernkopf |
Acoustic Echo Cancellation with Cross-Domain Learning (opens in new tab) |
4 |
12 |
Technische Universita ̈t Braunschweig |
Ernst Seidel, Jan Franzen, Maximilian Strake and Tim Fingscheidt |
Y^2-Net FCRN for Acoustic Echo and Noise Suppression (opens in new tab) |
5 |
14 |
Technion |
Amir Ivry, Israel Cohen, Baruch Berdugo |
Nonlinear Acoustic Echo Cancellation with Deep Learning (opens in new tab) |
NA |
NA |
NA |
Ross Cutler, Ando Saabas, Tanel Parnamaa, Markus Loide, Sten Sootla, Marju Purin, Hannes Gamper, Sebastian Braun, Karsten Sorensen, Robert Aichner and Sriram Srinivasan |
INTERSPEECH 2021 Acoustic Echo Cancellation Challenge (opens in new tab) |
Challenge submissions
Id | Team | ST NE MOS | ST FE Echo DMOS | DT Echo DMOS | DT Other DMOS | mean | CI |
4 | ERCESI | 4.25 | 4.59 | 4.69 | 4.18 | 4.43 | 0.02 |
2 | Trident | 4.27 | 4.49 | 4.52 | 4.39 | 4.42 | 0.02 |
7 | Kuaishou | 4.10 | 4.54 | 4.77 | 4.24 | 4.41 | 0.02 |
8 | ByteDance SAMI | 4.32 | 4.45 | 4.59 | 4.28 | 4.41 | 0.02 |
15 | Cruse run555 | 4.22 | 4.50 | 4.68 | 4.23 | 4.41 | 0.02 |
14 | Alibaba Group | 4.19 | 4.49 | 4.58 | 4.27 | 4.38 | 0.02 |
13 | Chinese Academy of Science | 4.26 | 4.34 | 4.36 | 4.23 | 4.30 | 0.02 |
5 | Bytedance | 4.23 | 4.49 | 4.31 | 4.15 | 4.29 | 0.02 |
15 | Tiny Cruse run2320 | 4.27 | 4.23 | 4.31 | 3.78 | 4.15 | 0.02 |
9 | Northwestern Polytechnical University | 3.78 | 4.44 | 4.44 | 3.90 | 4.14 | 0.02 |
11 | proactivaudio GmbH | 4.13 | 4.12 | 4.18 | 4.04 | 4.12 | 0.02 |
3 | Evolve Technologies | 4.01 | 4.52 | 3.90 | 3.72 | 4.04 | 0.02 |
15 | Baseline | 4.18 | 3.82 | 4.04 | 3.45 | 3.87 | 0.02 |
10 | Technische Universität Braunschweig | 4.16 | 3.73 | 3.72 | 3.53 | 3.78 | 0.03 |
15 | VQE MS + Tfnetv6 | 3.99 | 3.79 | 4.37 | 2.91 | 3.76 | 0.02 |
15 | VQE Skype + Tfnetv6 | 4.04 | 3.97 | 4.45 | 2.29 | 3.69 | 0.02 |
15 | VQE MS | 3.99 | 3.56 | 4.06 | 3.09 | 3.67 | 0.03 |
15 | VQE Skype | 3.95 | 3.87 | 4.31 | 2.40 | 3.63 | 0.02 |
12 | Merry Electronics | 3.29 | 3.83 | 4.21 | 2.92 | 3.56 | 0.03 |
6 | Technion | 2.73 | 2.50 | 3.53 | 3.40 | 3.04 | 0.03 |
15 | Universitat Politècncia de València | 2.25 | 3.37 | 3.76 | 1.92 | 2.82 | 0.03 |
Anova table of top entries
Team | ERCESI | Trident | Kuaishou | ByteDance SAMI |
ERCESI | 1 | 0.81 | 0.73 | 0.64 |
Trident | 0.81 | 1 | 0.91 | 0.82 |
Kuaishou | 0.73 | 0.91 | 1 | 0.91 |
ByteDance SAMI | 0.64 | 0.82 | 0.91 | 1 |
Legend
ST NE MOS: P.808 MOS of nearend singletalk scenario
ST FE Echo MOS: P.831 Echo DMOS for farend singletalk
DT Echo DMOS: P.831 Echo DMOS for doubletalk scenario
DT Other DMOS: P.831 other degradations DMOS of doubletalk scenario