AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

Adversarial Meta Sampling for Multilingual Low-Resource Speech Recognition

Yubei Xiao, Ke Gong, Pan Zhou, Guolin Zheng, Xiaodan Liang, Liang Lin

AAAI 2021 · Label-efficient and robust learning › Recognition and long-tail · approach: Multilingual language representation pretraining · 32 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Add adversarial meta sampling to MML-ASRMML-ASR · Diversity11, Indo9, and Indo12 target languagesWER (%): 58.29 -> 50.72 (Kyrgyz); 79.66 -> 72.26 (Estonian)Improved
Add adversarial meta sampling to MML-ASRMML-ASR · IARPA BABEL: Vietnamese, Swahili, and TamilCER (%): 45.10 -> 43.35; 36.14 -> 32.19; 50.61 -> 48.56Improved
Add adversarial meta sampling to MTL-ASRMTL-ASR · Diversity11, Indo9, and Indo12WER (%): 64.90 -> 59.55 (Kyrgyz); 83.70 -> 79.33 (Estonian); 62.55 -> 58.22 (Indo9 Dutch); 84.25 -> 83.90 (Indo9 Kabyle)Mixed
Add attention to the AMS policy network (ablation)AMS-MAML · Diversity11: KyrgyzWER (%): 54.16 -> 50.72Improved
Use learned adversarial sampling instead of uniform task sampling (ablation)MML-ASR (MAML) · Diversity11: Kyrgyz and EstonianWER (%): 58.29 -> 50.72; 79.66 -> 72.26Improved
Use learned adversarial sampling instead of probability proportional to task quantity (PPQ) (ablation)MML-ASR (MAML) · Diversity11: Kyrgyz and EstonianWER (%): 58.95 -> 50.72; 77.26 -> 72.26Improved
Use learned adversarial sampling instead of query-loss-proportional sampling (PPQL) (ablation)MML-ASR (MAML) · Diversity11: Kyrgyz and EstonianWER (%): 54.87 -> 50.72; 74.97 -> 72.26Improved
Use 80% rather than full target-language training data (ablation)AMS-MAML · Diversity11: KyrgyzWER (%): 50.72 -> 59.02Hurt
Use 50% rather than full target-language training data (ablation)AMS-MAML · Diversity11: KyrgyzWER (%): 50.72 -> 70.27Hurt
Use 20% rather than full target-language training data (ablation)AMS-MAML · Diversity11: KyrgyzWER (%): 50.72 -> 87.11Hurt
Apply AMS to FOMAMLFOMAML · Speech classification, average across D1–D5accuracy (%): 37.20 -> 39.63Improved
Apply AMS to FOMAMLFOMAML · Speech translation: Mongolian, Swedish, and TurkishBLEU: 0.35 -> 0.36; 0.67 -> 0.70; 1.26 -> 1.45Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy