Research map / Label-efficient and robust learning / Recognition and long-tail
Star Temporal Classification: Sequence Modeling with Partially Labeled Data
Vineel Pratap, Awni Hannun, Gabriel Synnaeve, Ronan Collobert
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 |
|---|---|---|---|
| Use STC instead of CTC on partially labeled data | Transformer · LibriSpeech p_drop=0.1 | test WER clean/other, greedy: 6.2/11.0 -> 4.1/9.0 | Improved |
| Use STC instead of CTC on partially labeled data | Transformer · LibriSpeech p_drop=0.4 | test WER clean/other, greedy: 48.2/56.9 -> 4.5/10.2 | Improved |
| Use STC instead of CTC on partially labeled data | Transformer · LibriSpeech p_drop=0.7 | test WER clean/other, greedy: 100/100 -> 6.9/15.1 | Improved |
| Use STC instead of CTC with p_drop varied by sample split | Transformer · LibriSpeech samples split in 3 | test WER clean/other, greedy: 58.6/63.9 -> 4.9/11.1 | Improved |
| Use STC instead of CTC with p_drop varied by word split | Transformer · LibriSpeech words split in 3 | test WER clean/other, greedy: 45.6/49.2 -> 5.8/13.0 | Improved |
| Add Transformer LM rescoring to STC decoding (ablation) | STC · LibriSpeech p_drop=0.1 | test WER clean/other: 4.1/9.0 -> 3.3/6.9 | Improved |
| Add 5-gram LM beam search to STC decoding (ablation) | STC · LibriSpeech p_drop=0.7 | test WER clean/other: 6.9/15.1 -> 7.0/14.8 | Mixed |
| Add pseudo labeling then CTC letter model | STC · LibriSpeech p_drop=0.1, Transformer rescoring | test WER clean/other: 3.3/6.9 -> 2.7/5.2 | Improved |
| Add pseudo labeling then CTC letter model | STC · LibriSpeech p_drop=0.7, Transformer rescoring | test WER clean/other: 5.1/11.1 -> 3.1/6.2 | Improved |
| Increase p_drop from 0.1 to 0.7 | STC · LibriSpeech greedy | test WER clean/other: 4.1/9.0 -> 6.9/15.1 | Hurt |
| Use STC instead of CTC | CNN · IAM p_drop=0.1 | test CER: 7.7 -> 7.2 | Improved |
| Use STC instead of CTC | CNN · IAM p_drop=0.7 | test CER: 78.5 -> 26.7 | Improved |
Closest papers by meaning
- W-CTC: a Connectionist Temporal Classification Loss with Wild Cards (ICLR 2022 · similarity 0.87)
- Variational Connectionist Temporal Classification (ECCV 2020 · similarity 0.86)
- Incremental Sequence Classification with Temporal Consistency (NeurIPS 2025 · similarity 0.85)
- Self-Distillation Regularized Connectionist Temporal Classification Loss for Text Recognition: A Simple Yet Effective Approach (AAAI 2024 · similarity 0.85)
- Align With Purpose: Optimize Desired Properties in CTC Models with a General Plug-and-Play Framework (ICLR 2024 · similarity 0.85)
- CR-CTC: Consistency regularization on CTC for improved speech recognition (ICLR 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.
