Research map / Architectures and efficiency / RNNs and normalization
FastCorrect: Fast Error Correction with Edit Alignment for Automatic Speech Recognition
Yichong Leng, Xu Tan, Linchen Zhu, Jin Xu, Renqian Luo, Linquan Liu, Tao Qin, Xiangyang Li, Edward Lin, Tie-Yan Liu
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 |
|---|---|---|---|
| remove edit alignment (use soft length-difference predictor) (ablation) | FastCorrect · AISHELL-1 test | WER (%): 4.16 -> 4.67 | Hurt |
| remove edit alignment (ablation) | FastCorrect · Internal dataset test | WER (%): 10.27 -> 12.27 | Hurt |
| remove pre-training on pseudo correction data (ablation) | FastCorrect · AISHELL-1 test | WER (%): 4.16 -> 4.83 | Hurt |
| remove pre-training (ablation) | FastCorrect · Internal dataset test | WER (%): 10.27 -> 10.33 | Hurt |
| remove pre-training (ablation) | AR model · AISHELL-1 test | WER (%): 4.08 -> 16.01 | Hurt |
| remove pre-training (ablation) | AR model · Internal dataset test | WER (%): 10.22 -> 10.26 | No clear effect |
| remove fine-tuning (pre-training only) (ablation) | FastCorrect · AISHELL-1 test | WER (%): 4.16 -> 5.19 | Hurt |
| remove fine-tuning (ablation) | FastCorrect · Internal dataset test | WER (%): 10.27 -> 11.74 | Hurt |
| use edit-alignment NAR FastCorrect instead of LevT | correction model · Internal dataset test | WER (%): 11.26 -> 10.27 | Improved |
| use FastCorrect instead of FELIX greedy alignment | correction model · AISHELL-1 test | WER (%): 4.63 -> 4.16 | Improved |
| use FastCorrect NAR instead of AR with deep encoder shallow decoder (10-2) | correction model · AISHELL-1 test | GPU latency (ms/sent): 84.0 -> 21.2 | Improved |
| use NAR FastCorrect instead of AR model | correction model · AISHELL-1 test | WER (%): 4.08 -> 4.16 | Hurt |
Closest papers by meaning
- Recording for Eyes, Not Echoing to Ears: Contextualized Spoken-to-Written Conversion of ASR Transcripts (AAAI 2025 · similarity 0.86)
- Speech Robust Bench: A Robustness Benchmark For Speech Recognition (ICLR 2025 · similarity 0.84)
- Speech Recognition Meets Large Language Model: Benchmarking, Models, and Exploration (AAAI 2025 · similarity 0.84)
- BIGOS V2 Benchmark for Polish ASR: Curated Datasets and Tools for Reproducible Evaluation (NeurIPS 2024 · similarity 0.84)
- Self-Taught Recognizer: Toward Unsupervised Adaptation for Speech Foundation Models (NeurIPS 2024 · similarity 0.84)
- ALRMR-GEC: Adjusting Learning Rate Based on Memory Rate to Optimize the Edit Scorer for Grammatical Error Correction (AAAI 2025 · 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.
