Research map / Label-efficient and robust learning / Self-supervised and contrastive
BEATs: Audio Pre-Training with Acoustic Tokenizers
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 |
|---|---|---|---|
| Replace reconstruction loss with masked discrete label prediction (random-projection tokenizer, iteration 1) | BEATs audio SSL model · AS-2M, ESC-50, KS2, ER | accuracy / mAP | Helped |
| Use self-distilled tokenizer (distilled from iter1 SSL teacher) instead of random-projection tokenizer (ablation) | BEATs iter2 · AS-20K | mAP: 36.0 to 38.3 | Helped |
| Use fine-tuned (AS-2M supervised) model as teacher for self-distilled tokenizer (ablation) | BEATs iter3+ · AS-2M | mAP: 48.0 to 48.6 | Helped |
| Use self-supervised teacher iter2 vs iter1 for tokenizer (iteration 3) (ablation) | BEATs iter3 · AS-2M | mAP: 48.1 to 48.0 | No clear effect |
| Use supervised fine-tuned model as tokenizer teacher instead of SSL teacher (ablation) | BEATs tokenizer · AS-20K, ESC-50 | mAP / accuracy: 38.3 to 38.9 (AS-20K) | Helped |
| Ensemble of 10 AS-2M fine-tuned BEATs models vs 5 | BEATs ensemble · AS-2M | mAP: 50.4 to 50.6 | Helped |
| Scale to larger model (Audio-MAE Large, 304M) compared with BEATs 90M is outperformed | baseline comparison · AS-2M | mAP | Mixed |
Closest papers by meaning
- SSLAM: Enhancing Self-Supervised Models with Audio Mixtures for Polyphonic Soundscapes (ICLR 2025 · similarity 0.88)
- MERT: Acoustic Music Understanding Model with Large-Scale Self-supervised Training (ICLR 2024 · similarity 0.87)
- ALMTokenizer: A Low-bitrate and Semantic-rich Audio Codec Tokenizer for Audio Language Modeling (ICML 2025 · similarity 0.87)
- WavTokenizer: an Efficient Acoustic Discrete Codec Tokenizer for Audio Language Modeling (ICLR 2025 · similarity 0.86)
- Weakly-supervised Audio Separation via Bi-modal Semantic Similarity (ICLR 2024 · similarity 0.86)
- CA-SSLR: Condition-Aware Self-Supervised Learning Representation for Generalized Speech Processing (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.