LSTMVAE and LSTMCVAE implemented with chainer.

This code is for Python 3. needs some Library as follows. chainer 2.0 cupy 1.0(if you use GPU) numpy>=1.12

Codes are based on Generating sentences from continuous space. written by Samuel R. Bowman in 2015.

Models can be downloaded from

VAE

. ./src/each_case/prof/model/download_model.sh

CVAE

. ./src/each_case/serif/model/download_model.sh

Test can be executed from

VAE

python ./src/each_case/sampleVAEProf.py

CVAE

python ./src/each_case/sampleCVAESerif.py

Hyperparameters are also set there.

As to VAE, training data are in ./src/each_case/prof/prof16000fixed_longcut.txt

As to CVAE, training data cannnot be uploaded due to the authority problems. Dummy data are in ./src/each_case/serif/all_serif16000_fixed_dummy.txt

and

./src/each_case/serif/all_chara_dummy.txt

Training can be executed from

VAE

python ./src/each_case/sampleVAEProf.py --train

CVAE

python ./src/each_case/sampleCVAESerif.py --train

Hyperparameters are also set there.

Then, test can be executed as follows.

VAE

python ./src/each_case/sampleVAEProf.py

CVAE

python ./src/each_case/sampleCVAESerif.py