Test and Manual Synthesis do not match

[JackTemaki]

Hello,

i just trained a model on the LJSpeech_1_1 dataset, and when i manually synthesize a sentence that was also part of the test set during training, i get slightly different results:

The original test alignment:

The alignment when using synthesize.py:

The later sound example also sounds worse in my opinion.

What could be the mistake i made?

best regards,
Nick

[JackTemaki]

Also, here are my commands and the hparams file:

python3 train.py --data_path=datasets/LJSpeech_1_1 --load_path=logs/LJSpeech_1_1_2018-12-17_15-04-28

python3 synthesizer.py --load_path=logs/LJSpeech_1_1_2018-12-17_15-04-28 --text "noyes was seized without difficulty, as it was a part of the scheme that he should act as the dupe, and remain on the spot in london till all the money was obtained." --is_korean=False

hparams.py.txt