Speech enhancement with variational autoencoders and alpha-stable distributions

Abstract

This paper focuses on single-channel semi-supervised speech enhancement. We learn a speaker-independent deep generative speech model using the framework of variational autoencoders. The noise model remains unsupervised because we do not assume prior knowledge of the noisy recording environment. In this context, our contribution is to propose a noise model based on alpha-stable distributions, instead of the more conventional Gaussian non-negative matrix factorization approach found in previous studies. We develop a Monte Carlo expectation-maximization algorithm for estimating the model parameters at test time. Experimental results show the superiority of the proposed approach both in terms of perceptual quality and intelligibility of the enhanced speech signal.

arXiv comment5 pages, 3 figures, audio examples and code available online : https://team.inria.fr/perception/research/icassp2019-asvae/. arXiv admin note: text overlap with arXiv:1811.06713

References in corpus (6)

• Adam: A Method for Stochastic Optimization
• Auto-Encoding Variational Bayes
• Supervised and Unsupervised Speech Enhancement Using Nonnegative Matrix Factorization
• Statistical Speech Enhancement Based on Probabilistic Integration of Variational Autoencoder and Non-Negative Matrix Factorization
• Semi-supervised multichannel speech enhancement with variational autoencoders and non-negative matrix factorization
• A variance modeling framework based on variational autoencoders for speech enhancement

Cited by in corpus (11)

• Audio-visual Speech Enhancement Using Conditional Variational Auto-Encoders
• Semi-supervised multichannel speech enhancement with variational autoencoders and non-negative matrix factorization
• Mixture of Inference Networks for VAE-based Audio-visual Speech Enhancement
• Learning and controlling the source-filter representation of speech with a variational autoencoder
• Generalized Fast Multichannel Nonnegative Matrix Factorization Based on Gaussian Scale Mixtures for Blind Source Separation
• Robust Speaker Recognition Using Speech Enhancement And Attention Model
• Fractional lower-order covariance-based measures for cyclostationary time series with heavy-tailed distributions: application to dependence testing and model order identification
• Speaker Re-identification with Speaker Dependent Speech Enhancement
• Deep Generative Variational Autoencoding for Replay Spoof Detection in Automatic Speaker Verification
• FastMVAE2: On improving and accelerating the fast variational autoencoder-based source separation algorithm for determined mixtures
• A Benchmark of Dynamical Variational Autoencoders applied to Speech Spectrogram Modeling