Speech Modeling with a Hierarchical Transformer Dynamical VAE
This work addresses speech processing tasks like enhancement by introducing a novel DVAE variant, though it is incremental as it builds on existing DVAE frameworks with architectural modifications.
The paper tackled the problem of modeling speech signals with dynamical variational autoencoders (DVAEs) by proposing HiT-DVAE, which uses a hierarchical Transformer architecture to replace recurrent neural networks for temporal dependencies, resulting in improved performance for speech spectrogram modeling and simpler training.
The dynamical variational autoencoders (DVAEs) are a family of latent-variable deep generative models that extends the VAE to model a sequence of observed data and a corresponding sequence of latent vectors. In almost all the DVAEs of the literature, the temporal dependencies within each sequence and across the two sequences are modeled with recurrent neural networks. In this paper, we propose to model speech signals with the Hierarchical Transformer DVAE (HiT-DVAE), which is a DVAE with two levels of latent variable (sequence-wise and frame-wise) and in which the temporal dependencies are implemented with the Transformer architecture. We show that HiT-DVAE outperforms several other DVAEs for speech spectrogram modeling, while enabling a simpler training procedure, revealing its high potential for downstream low-level speech processing tasks such as speech enhancement.