ASMR_Dataset / README.md
OOPPEENN's picture
Update README.md
2d74c18 verified
---
license: gpl-3.0
---
crawl progress: 57% 121249/213330
pipeline:
1. Manual cleaning, The following situations will be discarded:
- no Chinese subtitles
- timestamps of each sentence in the subtitles are connected at the end
- timestamps are not aligned
2. [mel band roformer](https://github.com/KimberleyJensen/Mel-Band-Roformer-Vocal-Model) remove SE, BGM, saliva sound, etc.
3. parsec subtitle & cut audio
4. Split the left and right channels and use [anime-whisper](https://huggingface.co/litagin/anime-whisper) to identify the loudest channel(beta version, maybe there is a better way)
5. create index