JEPA as a Neural Tokenizer: Learning Robust Speech Representations with Density Adaptive Attention

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Ioannides, Georgios, Constantinou, Christos, Chadha, Aman, Elkins, Aaron, Pang, Linsey, Shwartz-Ziv, Ravid, LeCun, Yann
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914186245701632
author Ioannides, Georgios
Constantinou, Christos
Chadha, Aman
Elkins, Aaron
Pang, Linsey
Shwartz-Ziv, Ravid
LeCun, Yann
author_facet Ioannides, Georgios
Constantinou, Christos
Chadha, Aman
Elkins, Aaron
Pang, Linsey
Shwartz-Ziv, Ravid
LeCun, Yann
contents We introduce a two-stage self-supervised framework that combines the Joint-Embedding Predictive Architecture (JEPA) with a Density Adaptive Attention Mechanism (DAAM) for learning robust speech representations. Stage~1 uses JEPA with DAAM to learn semantic audio features via masked prediction in latent space, fully decoupled from waveform reconstruction. Stage~2 leverages these representations for efficient tokenization using Finite Scalar Quantization (FSQ) and a mixed-radix packing scheme, followed by high-fidelity waveform reconstruction with a HiFi-GAN decoder. By integrating Gaussian mixture-based density-adaptive gating into the JEPA encoder, the model performs adaptive temporal feature selection and discovers hierarchical speech structure at a low frame rate of 2.5~Hz. The resulting tokens (47.5 tokens/sec) provide a reversible, highly compressed, and language-model-friendly representation that is competitive with, and often more efficient than, existing neural audio codecs.
format Preprint
id arxiv_https___arxiv_org_abs_2512_07168
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle JEPA as a Neural Tokenizer: Learning Robust Speech Representations with Density Adaptive Attention
Ioannides, Georgios
Constantinou, Christos
Chadha, Aman
Elkins, Aaron
Pang, Linsey
Shwartz-Ziv, Ravid
LeCun, Yann
Sound
Artificial Intelligence
Machine Learning
Audio and Speech Processing
We introduce a two-stage self-supervised framework that combines the Joint-Embedding Predictive Architecture (JEPA) with a Density Adaptive Attention Mechanism (DAAM) for learning robust speech representations. Stage~1 uses JEPA with DAAM to learn semantic audio features via masked prediction in latent space, fully decoupled from waveform reconstruction. Stage~2 leverages these representations for efficient tokenization using Finite Scalar Quantization (FSQ) and a mixed-radix packing scheme, followed by high-fidelity waveform reconstruction with a HiFi-GAN decoder. By integrating Gaussian mixture-based density-adaptive gating into the JEPA encoder, the model performs adaptive temporal feature selection and discovers hierarchical speech structure at a low frame rate of 2.5~Hz. The resulting tokens (47.5 tokens/sec) provide a reversible, highly compressed, and language-model-friendly representation that is competitive with, and often more efficient than, existing neural audio codecs.
title JEPA as a Neural Tokenizer: Learning Robust Speech Representations with Density Adaptive Attention
topic Sound
Artificial Intelligence
Machine Learning
Audio and Speech Processing
url https://arxiv.org/abs/2512.07168