A weighted-variance variational autoencoder model for speech enhancement
This work addresses speech enhancement for audio processing applications, presenting an incremental improvement over existing variational autoencoder methods.
The paper tackles speech enhancement by proposing a weighted-variance variational autoencoder model that uses a Gamma prior on weights to achieve a Student's t-distribution for speech generative modeling, showing effectiveness and robustness in experiments compared to standard unweighted models.
We address speech enhancement based on variational autoencoders, which involves learning a speech prior distribution in the time-frequency (TF) domain. A zero-mean complex-valued Gaussian distribution is usually assumed for the generative model, where the speech information is encoded in the variance as a function of a latent variable. In contrast to this commonly used approach, we propose a weighted variance generative model, where the contribution of each spectrogram time-frame in parameter learning is weighted. We impose a Gamma prior distribution on the weights, which would effectively lead to a Student's t-distribution instead of Gaussian for speech generative modeling. We develop efficient training and speech enhancement algorithms based on the proposed generative model. Our experimental results on spectrogram auto-encoding and speech enhancement demonstrate the effectiveness and robustness of the proposed approach compared to the standard unweighted variance model.