How to start

first, you need to have python 3 installed along with Tensorflow v1.6.

next you can install the requirements using:

pip install -r requirements.txt

Basic Dataset:

We tested the code above on the ljspeech dataset, which has almost 24 hours of labeled single actress voice recording. (further info on the dataset are available in the README file when you download it)

After downloading the dataset, extract the compressed file, and place the folder inside the cloned repository.

Preprocessing

From this point and further, you'll have to be located inside the "tacotron" folder

cd tacotron

Preprocessing can then be started using:

python preprocess.py

This should take few minutes.

Training:

Feature prediction model can be trained using:

python train.py

checkpoints will be made each 100 steps and stored under logs-<model_name> folder.

Synthesis

There are three types of mel spectrograms synthesis using this model:

• Evaluation (synthesis on custom sentences). This is what we'll usually use after having a full end to end model.

python synthesize.py --mode='eval'

• Natural synthesis (let the model make predictions alone by feeding last decoder output to the next time step).

python synthesize.py --GTA=False

• Ground Truth Aligned synthesis (DEFAULT: the model is assisted by true labels in a teacher forcing manner). This synthesis method is used when predicting mel spectrograms used to train the wavenet vocoder. (yields better results as stated in the paper)

python synthesize.py

References:

https://github.com/Rayhane-mamah/Tacotron-2