Good Work, trained with LJSpeech and 23 hours,

[icewwn]

Amazing work, trained with LJSpeech on 1070 Ti in 23 hours,
I got the good synthesized samples on ground truth

Above is recordings,
Below is synthesized by synthesis.py with ground truth.
Sound good except end noise on silence segment.

[r9y9]

Hi, thanks. I think you just need to be patient. From my experience, it usually requires more than 1000k steps (~2 weeks on a single 1080Ti) to get noise-less high-quality synthetic speech.