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
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 changed | Where | Measured | Effect |
|---|---|---|---|
| Switch from local to global normalization | Streaming GNAT, 1-gram · LibriSpeech test_clean | WER: 7.0 -> 5.5 | Improved |
| Switch from local to global normalization | Streaming GNAT, 1-gram · LibriSpeech test_other | WER: 17.4 -> 14.0 | Improved |
| Switch from local to global normalization | Streaming GNAT, 2-gram · LibriSpeech test_clean | WER: 4.9 -> 3.8 | Improved |
| Switch from local to global normalization | Streaming GNAT, 2-gram · LibriSpeech test_other | WER: 11.0 -> 9.5 | Improved |
| Switch from local to global normalization | Non-streaming GNAT, 1-gram · LibriSpeech test_clean | WER: 3.4 -> 3.3 | No clear effect |
| Switch from local to global normalization | Non-streaming GNAT, 2-gram · LibriSpeech test_clean | WER: 2.8 -> 2.8 | No clear effect |
| Increase context dependency from 1-gram to 2-gram (ablation) | Streaming GNAT, global · LibriSpeech test_clean | WER: 5.5 -> 3.8 | Improved |
| Increase context dependency from 1-gram to 2-gram (ablation) | Streaming GNAT, global · LibriSpeech test_other | WER: 14.0 -> 9.5 | Improved |
| Use shared-rnn instead of shared-emb weight function (ablation) | Streaming GNAT, 2-gram, global · LibriSpeech test_clean | WER: 4.1 -> 3.8 | Improved |
| Use shared-rnn instead of shared-emb weight function (ablation) | Streaming GNAT, 2-gram, global · LibriSpeech test_other | WER: 9.9 -> 9.5 | Improved |
| Use shared-rnn instead of unshared weight function (ablation) | Streaming GNAT, 2-gram, global · LibriSpeech test_clean | WER: 4.2 -> 3.8 | Improved |
| Use shared-rnn instead of unshared weight function (ablation) | Streaming GNAT, 2-gram, global · LibriSpeech test_other | WER: 10.6 -> 9.5 | Improved |
Closest papers by meaning
- Unified Speech Recognition: A Single Model for Auditory, Visual, and Audiovisual Inputs (NeurIPS 2024 · similarity 0.85)
- Speech Recognition Meets Large Language Model: Benchmarking, Models, and Exploration (AAAI 2025 · similarity 0.85)
- HAINAN: Fast and Accurate Transducer for Hybrid-Autoregressive ASR (ICLR 2025 · similarity 0.85)
- Heuristic-free Knowledge Distillation for Streaming ASR via Multi-modal Training (AAAI 2025 · similarity 0.84)
- DiffNorm: Self-Supervised Normalization for Non-autoregressive Speech-to-speech Translation (NeurIPS 2024 · similarity 0.84)
- Self-Taught Recognizer: Toward Unsupervised Adaptation for Speech Foundation Models (NeurIPS 2024 · similarity 0.84)
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.
