Article Dans Une Revue IEEE Signal Processing Letters Année : 2025

Posterior Transition Modeling for Unsupervised Diffusion-Based Speech Enhancement

Résumé

We explore unsupervised speech enhancement using diffusion models as expressive generative priors for clean speech. Existing approaches guide the reverse diffusion process using noisy speech through an approximate, noise-perturbed likelihood score, combined with the unconditional score via a trade-off hyperparameter. In this work, we propose two alternative algorithms that directly model the conditional reverse transition distribution of diffusion states. The first method integrates the diffusion prior with the observation model in a principled way, removing the need for hyperparameter tuning. The second defines a diffusion process over the noisy speech itself, yielding a fully tractable and exact likelihood score. Experiments on the WSJ0-QUT and VoiceBank-DEMAND datasets demonstrate improved enhancement metrics and greater robustness to domain shifts compared to both supervised and unsupervised baselines.

Fichier principal
Vignette du fichier
qvjhqxqmgphmgppfkccpcbdhxdrnnbjz.pdf (355.44 Ko) Télécharger le fichier
Origine Fichiers produits par l'(les) auteur(s)
Licence

Dates et versions

hal-05135495 , version 1 (30-06-2025)

Licence

Identifiants

Citer

Mostafa Sadeghi, Jean-Eudes Ayilo, Romain Serizel, Xavier Alameda-Pineda. Posterior Transition Modeling for Unsupervised Diffusion-Based Speech Enhancement. IEEE Signal Processing Letters, 2025, 32, pp.2694-2698. ⟨10.1109/LSP.2025.3583967⟩. ⟨hal-05135495⟩
347 Consultations
185 Téléchargements

Altmetric

Partager

  • More