search

rhasspy / piper

A fast, local neural text to speech system
https://rhasspy.github.io/piper-samples/
MIT License
6.28k stars 461 forks source link

train fails with LJSpeech-1.1 #10

Closed igorsitdikov closed 1 year ago

igorsitdikov commented 1 year ago

When run train it fails with error assert utt.speaker_id is not None, "Missing speaker id" AssertionError: Missing speaker id I thought that it should work on dataset which you've added in README. dataset was prepared with scripts which is provided in README.

synesthesiam commented 1 year ago

I think there was a bug in the preprocessing script that should be fixed. Try editing your config.json file and changing num_speakers to 1.

igorsitdikov commented 1 year ago

Thank you, it works. Is there way to continue train from checkpoint? could you please provide command?

synesthesiam commented 1 year ago

Yes, add --resume_from_checkpoint <path> to the training command. You can see all of the options with --help

igorsitdikov commented 1 year ago

Thanks a lot!