AI Paper Atlas

Research map / Label-efficient and robust learning / Self-supervised and contrastive

Variable-rate hierarchical CPC leads to acoustic unit discovery in speech

Santiago Cuervo, Adrian Lancucki, Ricard Marxer, Paweł Rychlikowski, Jan K Chorowski

NeurIPS 2022 · Label-efficient and robust learning › Self-supervised and contrastive · approach: Self-supervised speech and brain decoding · 3 citations (OpenAlex)

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 changedWhereMeasuredEffect
Use variable-rate downsampling learned from the high-level CPC objective instead of no downsamplingTwo-level CPC · LibriSpeech clean 100 testFrame accuracy: 67.49 -> 72.57Improved
Use variable-rate downsampling instead of fixed-rate downsamplingTwo-level CPC · LibriSpeech clean 100 testFrame accuracy: 67.92 -> 72.57Improved
Add high-level CPC supervision with learned variable-rate segmentationFrame-level CPC representations · LibriSpeech clean 100 testFrame accuracy: 68.60 -> 72.57Improved
Use variable-rate segmentation rather than phone-label-supervised downsamplingFrame-level CPC representations · LibriSpeech clean 100 testFrame accuracy: 71.01 -> 72.57Improved
Use variable-rate learned downsampling rather than no downsamplingHigh-level CPC · LibriSpeech clean 100 testPhone accuracy: 83.41 -> 78.93Hurt
Use phone-label-supervised downsampling rather than learned variable-rate downsamplingHigh-level CPC · LibriSpeech clean 100 testPhone accuracy: 78.93 -> 85.74Improved
Use the proposed segmentation instead of SCPC segmentationPhone boundary detection · LibriSpeech clean 100 testPrecision: 64.05 -> 79.94Improved
Use the proposed segmentation instead of SCPC segmentationPhone boundary detection · TIMIT test (non-speech removed)F1: 85.31 -> 80.73Hurt
Add adjacent negative sampling to enforce distinguishability of successive units (ablation)Full hierarchical CPC · LibriSpeech clean 100 testFrame accuracy: 64.13 -> 72.52Improved
Add adjacent negative sampling to enforce distinguishability of successive units (ablation)Full hierarchical CPC · LibriSpeech clean 100 testR-val: 21.10 -> 81.09Improved
Add average segment-length policy regularization (ablation)Hierarchical CPC with adjacent negative sampling · LibriSpeech clean 100 testFrame accuracy: 66.87 -> 72.52Improved
Add average segment-length policy regularization (ablation)Hierarchical CPC with adjacent negative sampling · LibriSpeech clean 100 testR-val: 0.0 -> 81.09Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy