Research map / Label-efficient and robust learning / Self-supervised and contrastive
Unsupervised Speech Recognition
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 wav2vec-U instead of the best previous unsupervised approach | Unsupervised speech recognition · TIMIT all-test, matched | PER: 26.1 -> 11.3 | Helped |
| Add self-training and a Transformer language model | wav2vec-U Large · Librispeech test-other | WER: 5.9 | Helped |
| Add self-training | wav2vec-U · Librispeech | WER | Helped |
| Use 960 hours rather than 53.2k hours of unlabeled speech audio for self-training | wav2vec-U · Librispeech dev-other | WER: 6.4 vs. 6.0 | Hurt |
| Use block 15 representations instead of other wav2vec 2.0 layers (ablation) | Frozen wav2vec 2.0 phoneme classifier · Librispeech dev-other | PER: 7.5 PER | Helped |
| Remove silence segments from audio (ablation) | wav2vec-U · Librispeech dev-other | PER | Helped |
| Add silence tokens at sentence boundaries in phonemized text (ablation) | wav2vec-U · Librispeech dev-other | PER | Helped |
| Insert silence tokens between words at a rate of 0.25 (ablation) | wav2vec-U · Librispeech dev-other | PER | Helped |
| Add a phonemic silence label (ablation) | wav2vec-U generator · Speech recognition | PER | Helped |
| Add segment smoothness and phoneme diversity penalties (ablation) | wav2vec-U adversarial objective · Speech recognition | PER | Helped |
| Add a second HMM self-training iteration (ablation) | wav2vec-U + HMM · TIMIT | PER | Helped |
| Resegment speech with an HMM and repeat GAN training (ablation) | wav2vec-U + HMM · TIMIT | PER | No clear effect |
Closest papers by meaning
- wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations (NeurIPS 2020 · similarity 0.90)
- Unsupervised Speech Recognition via Segmental Empirical Output Distribution Matching (ICLR 2019 · similarity 0.90)
- Unified Speech Recognition: A Single Model for Auditory, Visual, and Audiovisual Inputs (NeurIPS 2024 · similarity 0.88)
- UniWav: Towards Unified Pre-training for Speech Representation Learning and Generation (ICLR 2025 · similarity 0.88)
- Self-Taught Recognizer: Toward Unsupervised Adaptation for Speech Foundation Models (NeurIPS 2024 · similarity 0.87)
- LAMA-UT: Language Agnostic Multilingual ASR Through Orthography Unification and Language-Specific Transliteration (AAAI 2025 · similarity 0.85)
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.