Implementation of tacotron, a text to speech deep learning model.
Paper can be found here: tacotron.
-
Dataset can be retrieved from here. When extracted, file will contain LJSpeech-1.1 as a folder. Move that folder into the root directory beside where train.py is.
-
Run preprocess.py to preprocess the audio files. Audio files will be generated in a directory called training for the default parameters.
-
After running the preprocessing steps, we can start training the model.
python train.py
todo.
TODO:
- train the network to generate a pretrained model
Reimplementation of this for education purposes.
Got lots of reference from: https://github.com/keithito/tacotron Really grateful and appreciate the work of Keith Ito.