EveryVoiceTTS / EveryVoiceTTS/EveryVoice
Investigate zipping preprocessed files to help save inodes
Open
enhancement
- Dominant language
- Python
- Stars
- 45
- Forks
- 4
- Avg merge
- 1d 8h
- Merged PRs (30d)
- 14
Description
@SamuelLarkin suggests fuse-mounting a zip archive or a new dataloader that preprocesses the audio on-the-fly (if it doesn't slowdown training aka makes the trainer wait on preprocessing) to help circumvent the perennial inode problem for large datasets.
Contributor guide
Assessment
This issue has not been assessed yet.