wgfi110 / autoKWS2021_1st_solution

Auto-KWS 2021 Challenge 1st place solution.

INTERSPEECH2021 Auto-KWS 2021 Challenge 冠军方案

---

赛题见如下链接

比赛宣传页:https://www.4paradigm.com/competition/autospeech2021
Feedback Phase(初赛): https://www.automl.ai/competitions/15#home
Final Phase(复赛): https://www.automl.ai/competitions/12#home

---

代码说明

主要代码是enrollment.sh和predict.sh。详细过程见脚本内注释。

最优结果由两种STD模型融合+一个说话人模型(ECAPA-TDNN)获得。使用单个CNN_QbE_STD模型+ECAPA-TDNN模型在Feedback Phase可获得0.457的分数。

---

Final Phase排行榜

---

致谢

感谢以下优秀的开源工作：
https://github.com/fauxneticien/bnf_cnn_qbe-std
https://github.com/idiap/CNN_QbE_STD
https://github.com/wenet-e2e/wenet
https://github.com/kaldi-asr/kaldi
https://huggingface.co/speechbrain/spkrec-ecapa-voxceleb
...

参考文献
[1] Dhananjay Ram, Lesly Miculicich, Herve Bourlard. CNN based Query by Example Spoken Term Detection. INTERSPEECH 2018.
[2] Brecht Desplanques, Jenthe Thienpondt, Kris Demuynck. ECAPA-TDNN: Emphasized Channel Attention, Propagation and Aggregation in TDNN Based Speaker Verification. INTERSPEECH 2020.