NVIDIA / radtts

Provides training, inference and voice conversion recipes for RADTTS and RADTTS++: Flow-based TTS models with Robust Alignment Learning, Diverse Synthesis, and Generative Modeling and Fine-Grained Control over of Low Dimensional (F0 and Energy) Speech Attributes.
MIT License
283 stars 40 forks source link

Certain texts in LJ speech unloadable #20

Closed sjkoelle closed 1 year ago

sjkoelle commented 1 year ago

I am getting an EOF error on certain data points within the LJ speech dataset. In particular, the text was after 1807, through the exertions of the keeper of the jail, spent in the purchase of necessaries. does not work, while both was after 1807 through the exertions of the keeper of the jail spent in the purchase of necessaries. and was after 1807, through the exertions of the keeper of the jail, spent in the purchase of necessaries do. Any idea why this is occurring and how I can fix this issue? Thanks

sjkoelle commented 1 year ago

After looking more closely at this issue, I think it is because certain beta binomial prior paths in data_cache are somehow corrupted. Removing the corrupted cached file and allowing it to be regenerated appears to fix the issue.