Research map / Label-efficient and robust learning / Self-supervised and contrastive
Self-supervised speech representations
Learns speech and audio representations with self-supervised unified frameworks and pruning.
24 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2020–2026) · an approach in Self-supervised and contrastive, keyword cluster “supervised ssl · representations · masked”
State History Tried Checklist Preprints Who Export
Where it stands
Not enough benchmark data Fewer than four papers report the same benchmark, so there is no record to follow.
9 papers in 2023–24 and 7 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 2.8% to 2.2%.
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2020 · vq-wav2vec: Self-Supervised Learning of Discrete Speech Representations (ICLR 2020 · 415 citations)
- 2021 · Task Agnostic and Task Specific Self-Supervised Learning from Speech with LeBenchmark (NeurIPS 2021 · 19 citations)
- 2022 · ContentVec: An Improved Self-Supervised Speech Representation by Disentangling Speakers (ICML 2022 · 24 citations)
- 2023 · BEATs: Audio Pre-Training with Acoustic Tokenizers (ICML 2023 · 41 citations)
- 2024 · MERT: Acoustic Music Understanding Model with Large-Scale Self-supervised Training (ICLR 2024 · 28 citations)
- 2025 · Closed-Form Training Dynamics Reveal Learned Features and Linear Structure in Word2Vec-like Models (NeurIPS 2025 · 0 citations)
- 2026 · BAT: Better Audio Transformer Guided by Convex Gated Probing (ICML 2026 · 0 citations)
Earliest papers here
- vq-wav2vec: Self-Supervised Learning of Discrete Speech Representations (ICLR 2020 · 415 citations)
- Task Agnostic and Task Specific Self-Supervised Learning from Speech with LeBenchmark (NeurIPS 2021 · 19 citations)
- PARP: Prune, Adjust and Re-Prune for Self-Supervised Speech Recognition (NeurIPS 2021 · 12 citations)
What has been tried
19 papers here report 227 measured changes (a component added, removed or swapped, compared with a baseline): helped 63%, no clear effect 10%, hurt 19%, mixed 7%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Add a CQT musical teacher to K-means Logmel+Chroma acoustic targets | Giantsteps key detection · refined accuracy | 55.1 -> 65.0 | Helped | MERT: Acoustic Music Understanding Model with Large-Scale Self-supervised Training (ICLR 2024 · 28 citations) |
| Add TCAC time-wise attention instead of channel-wise conditioning | ML-SUPERB 10-min, Normal and Few-shots · ASR CER | 18.6 -> 17.8; 32.2 -> 31.8 | Helped | CA-SSLR: Condition-Aware Self-Supervised Learning Representation for Generalized Speech Processing (NeurIPS 2024 · 2 citations) |
| add a third resolution (20, 40, 80 ms) | LibriSpeech 10-hour labeled, test-clean · WER | 8.46 -> 8.84 | Hurt | Multi-resolution HuBERT: Multi-resolution Speech Self-Supervised Learning with Masked Unit Prediction (ICLR 2024 · 1 citation) |
| Use the quartic loss approximation instead of the vanilla word2vec loss | COCA and NOW mixture · singular value dynamics | No clear effect | Closed-Form Training Dynamics Reveal Learned Features and Linear Structure in Word2Vec-like Models (NeurIPS 2025 · 0 citations) | |
| use Sylber greedy segmentation instead of MinCut | LibriSpeech test · F1 | 72.2 -> 72.2 | No clear effect | Sylber: Syllabic Embedding Representation of Speech from Raw Audio (ICLR 2025 · 0 citations) |
| add partially mixed audio objectives instead of unmixed-only objectives | SPASS and other polyphonic datasets · mAP | 62.8 -> 67.4 | Helped | SSLAM: Enhancing Self-Supervised Models with Audio Mixtures for Polyphonic Soundscapes (ICLR 2025 · 0 citations) |
| add heuristic input transformations | LibriTTS test-clean zero-shot voice conversion · SV-EER | 28.9% -> 6.0% | Helped | SelfVC: Voice Conversion With Iterative Refinement using Self Transformations (ICML 2024 · 0 citations) |
| Add novel speaker combinations to synthetic speech | English ASR, 100-hour real speech · CER | 6.91 -> 5.19 | Helped | Speech Self-Supervised Learning Using Diffusion Model Synthetic Data (ICML 2024) |
What reviewers will expect
Datasets and baselines named in the 9 papers of this approach from 2024–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- LibriLight · 3 of 9 papers
- LibriSpeech · 2 of 9 papers
- ML-SUPERB · 2 of 9 papers
Baselines
- WAVLM · 2 of 9 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 16 papers; from OpenAlex and the papers' affiliations).
- Carnegie Mellon University · 2 papers
- Meta · 2 papers
Use it
Download BibTeX of 24 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Invariance-based self-supervised learning (168 papers)
- Data-efficient and robust self-supervision (122 papers)
- Masked autoencoder pretraining (48 papers)
Elsewhere in the topic
- Pretext-task visual pre-training (136 papers)
- Self-supervision in supervised pipelines (121 papers)
- Self-supervised neuroimaging representations (98 papers)
- Contrastive versus non-contrastive theory (74 papers)
- Inductive bias and minimal representations (66 papers)
- Self-supervised speech and brain decoding (56 papers)
- Empirical contrastive visual learning (38 papers)
- Representation evaluation and trade-offs (37 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).