Learning Dependencies of Discrete Speech Representations with Neural Hidden Markov Models
This work addresses the need for better phonetic modeling in self-supervised speech processing, offering incremental improvements over existing methods.
The paper tackled the problem of modeling dependencies among discrete latent variables in speech representation learning, proposing neural hidden Markov models to capture frame-level Markovian dependencies, which improved phonetic information accessibility, segmentation, and cluster purity compared to independent-frame models.
While discrete latent variable models have had great success in self-supervised learning, most models assume that frames are independent. Due to the segmental nature of phonemes in speech perception, modeling dependencies among latent variables at the frame level can potentially improve the learned representations on phonetic-related tasks. In this work, we assume Markovian dependencies among latent variables, and propose to learn speech representations with neural hidden Markov models. Our general framework allows us to compare to self-supervised models that assume independence, while keeping the number of parameters fixed. The added dependencies improve the accessibility of phonetic information, phonetic segmentation, and the cluster purity of phones, showcasing the benefit of the assumed dependencies.