LL-SDR: Low-Latency Speech enhancement through Discrete Representations

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Li, Jingyi, Della Libera, Luca, Ravanelli, Mirco, Subakan, Cem
Formato: Preprint
Publicado: 2026
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866912976409198592
author Li, Jingyi
Della Libera, Luca
Ravanelli, Mirco
Subakan, Cem
author_facet Li, Jingyi
Della Libera, Luca
Ravanelli, Mirco
Subakan, Cem
contents Many speech enhancement (SE) methods rely on continuous representations. Recently, discrete audio tokens have been explored to enable autoregressive generation for SE. However, it remains unclear whether discretization itself consistently improves SE performance. In this paper, we introduce LL-SDR, a token-based speech enhancement framework that explicitly leverages discretization to better separate speech and noise. Our first contribution is a Variance-Ordered Residual Vector Quantizer (VO-RVQ), designed to disentangle speech and noise distributions during tokenization. Second, we propose a latent-space discriminator to better align enhanced embeddings with semantic embeddings. Experiments show that LL-SDR outperforms continuous baselines and matches the performance of autoregressive token-based approaches, while enabling lightweight, low-latency speech enhancement in both reverberant and non-reverberant noisy environments. Demos and source code are available at our project websites.
format Preprint
id arxiv_https___arxiv_org_abs_2603_20242
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle LL-SDR: Low-Latency Speech enhancement through Discrete Representations
Li, Jingyi
Della Libera, Luca
Ravanelli, Mirco
Subakan, Cem
Sound
Audio and Speech Processing
Many speech enhancement (SE) methods rely on continuous representations. Recently, discrete audio tokens have been explored to enable autoregressive generation for SE. However, it remains unclear whether discretization itself consistently improves SE performance. In this paper, we introduce LL-SDR, a token-based speech enhancement framework that explicitly leverages discretization to better separate speech and noise. Our first contribution is a Variance-Ordered Residual Vector Quantizer (VO-RVQ), designed to disentangle speech and noise distributions during tokenization. Second, we propose a latent-space discriminator to better align enhanced embeddings with semantic embeddings. Experiments show that LL-SDR outperforms continuous baselines and matches the performance of autoregressive token-based approaches, while enabling lightweight, low-latency speech enhancement in both reverberant and non-reverberant noisy environments. Demos and source code are available at our project websites.
title LL-SDR: Low-Latency Speech enhancement through Discrete Representations
topic Sound
Audio and Speech Processing
url https://arxiv.org/abs/2603.20242