Greatful for data contribution from Mr. Meekit Intarapipat, senior Thai musician (Ranat Ek)
This projected contribute to AIB
We have deployed Variational AutoEncoder or VAE to regenerate Ranat Eak Thai classical music. Trained on 45 hours dataset, we have choosen the 3 best of all models.
This Project base on Tensorflow 2.4.0 framework
links to the models asset
Master_CNN: architecture implemented in 4 layers CNN-only VAE https://drive.google.com/file/d/1yF6MPZpW23rvICCHo9Tht7QVrwpTVzKK/view?usp=sharing
Master_CNN_LSTM: architecture with LSTM instead of Flatten https://drive.google.com/file/d/1D6J5O-Bz0tSapaEgO_iXa2S2MjzBaS4p/view?usp=sharing
Wierdo's architecture: repeat CNN followed by LSTM https://drive.google.com/file/d/1RNtRNTWWej4d_PV9VZyuiYzQid4-ynPP/view?usp=sharing