AI Paper Atlas

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

Global Normalization for Streaming Speech Recognition in a Modular Framework

Ehsan Variani, Ke Wu, Michael Riley, David Rybach, Matt Shannon, Cyril Allauzen

NeurIPS 2022 · Label-efficient and robust learning › Recognition and long-tail · approach: Multilingual language representation pretraining · 6 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
Switch from local to global normalizationStreaming GNAT, 1-gram · LibriSpeech test_cleanWER: 7.0 -> 5.5Improved
Switch from local to global normalizationStreaming GNAT, 1-gram · LibriSpeech test_otherWER: 17.4 -> 14.0Improved
Switch from local to global normalizationStreaming GNAT, 2-gram · LibriSpeech test_cleanWER: 4.9 -> 3.8Improved
Switch from local to global normalizationStreaming GNAT, 2-gram · LibriSpeech test_otherWER: 11.0 -> 9.5Improved
Switch from local to global normalizationNon-streaming GNAT, 1-gram · LibriSpeech test_cleanWER: 3.4 -> 3.3No clear effect
Switch from local to global normalizationNon-streaming GNAT, 2-gram · LibriSpeech test_cleanWER: 2.8 -> 2.8No clear effect
Increase context dependency from 1-gram to 2-gram (ablation)Streaming GNAT, global · LibriSpeech test_cleanWER: 5.5 -> 3.8Improved
Increase context dependency from 1-gram to 2-gram (ablation)Streaming GNAT, global · LibriSpeech test_otherWER: 14.0 -> 9.5Improved
Use shared-rnn instead of shared-emb weight function (ablation)Streaming GNAT, 2-gram, global · LibriSpeech test_cleanWER: 4.1 -> 3.8Improved
Use shared-rnn instead of shared-emb weight function (ablation)Streaming GNAT, 2-gram, global · LibriSpeech test_otherWER: 9.9 -> 9.5Improved
Use shared-rnn instead of unshared weight function (ablation)Streaming GNAT, 2-gram, global · LibriSpeech test_cleanWER: 4.2 -> 3.8Improved
Use shared-rnn instead of unshared weight function (ablation)Streaming GNAT, 2-gram, global · LibriSpeech test_otherWER: 10.6 -> 9.5Improved

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