Open margaritatsobenko opened 2 years ago
To improve Tacotron's results (to minimize variation of confidence interval of scores) try to change hyperparameters (convolution params, LSTM params...). NB: It's better to have the same params for each convolution layer.
Models: