After the previous steps, training and testing can be achieved through 3_train.py and 4_test.py.
The first script trains a Tacotron model on the prepared training set across the NB_EPOCHS epochs, and then saves the model in the /results folder.
The second script allows the user to apply the previously saved model on any transcript of testing dataset. The selection of the audio to predict is done through a variable, item_index, which should contain the index (in the testing dataset) of the wanted item.
The estimated spectrogram is then converted to a waveform through the Griffin-Lim algorithm. The conversion function, from_spectro_to_waveform, is defined in the /processing/proc_audio.py file.
We strongly encourage ...