what are the characteristics or basis for other split
#3
by
Lhma-aslp
- opened
Great work! I would like to know what are the characteristics or basis for other split in the Data Studio ? How is it different from the train split ?
The other split is created to exclude audio files unsuitable for fine-tuning Cosyvoice. This split contains:
- Missing vocalizations: Files lacking Nonverbal Vocalizations after annotation.
- Speaker overlap: Files from speakers already in the validation/test splits to prevent contamination of the training data.
Thanks for the clear explanation! Really appreciate your help!
Lhma-aslp
changed discussion status to
closed