Déréverbération non-supervisée de la parole par modèle hybride

2025-10-10

AI Summary

This paper addresses the fundamental challenge in unsupervised speech dereverberation: the absence of paired clean/reverberant speech data. We propose a novel training paradigm that relies solely on reverberant speech and a small set of readily available acoustic parameters—most notably RT60. Methodologically, we design a hybrid neural network architecture explicitly incorporating RT60 priors and formulate an end-to-end unsupervised objective via self-supervised signal modeling, eliminating the need for clean-speech supervision. Our key contribution lies in the explicit integration of sparse yet easily measurable reverberation parameters into the training process, thereby enhancing generalization to unseen reverberant conditions and improving robustness. Experiments demonstrate that our approach significantly outperforms existing state-of-the-art unsupervised methods across standard metrics—including PESQ and STOI—and exhibits superior cross-dataset generalization stability.

Technology Category

• Machine Learning: Unsupervised & Self-Supervised Learning
• Natural Language Processing: Speech
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs

Abstract

This paper introduces a new training strategy to improve speech dereverberation systems in an unsupervised manner using only reverberant speech. Most existing algorithms rely on paired dry/reverberant data, which is difficult to obtain. Our approach uses limited acoustic information, like the reverberation time (RT60), to train a dereverberation system. Experimental results demonstrate that our method achieves more consistent performance across various objective metrics than the state-of-the-art.

Problem

• Developing unsupervised speech dereverberation without clean data
• Overcoming reliance on paired dry and reverberant speech data
• Training systems using limited acoustic information like RT60

Innovation

• Unsupervised training strategy for speech dereverberation
• Uses only reverberant speech without paired data
• Employs limited acoustic information like RT60