At first it all worked well, after some time, it started generating very weird outputs. This is my command:

./tts --text "Hello there" --out_path ~/hello-there.wav

and I get the file sounding like a ghost, hear it please, and tell me why is this happening?
hello-there.webm

I have used standard model tts_models/en/ljspeech/tacotron2-DDC and I do not see this effect when I am using Thorsten German language model. I think it is model related.