Re-examination of the Role of Latent Variables in Sequence Modeling

This repository contains the code in PyTorch for our paper

Re-examination of the Role of Latent Variables in Sequence Modeling

Zihang Dai*, Guokun Lai*, Yiming Yang, Shinjae Yoo (*: equal contribution)

NeurIPS 2019

Replicating results in the paper

For each type of datasets (speech, midi, handwriting), please refer to the README.md in the corresponding directory in the repo for detailed instructions for replicating our work.

Datasets considered in this work

speech datasets:

• TIMIT
• VCTK
• Blizzard

midi music datasets:

• MUSE
• Nottingham

handwriting trajectory dataset:

• IAM-OnDB