| --- |
| license: openrail |
| language: |
| - en |
| - ja |
| thumbnail: >- |
| https://cdn.discordapp.com/attachments/660770162072485890/1132232063349686303/HarutyaRVC.png |
| tags: |
| - harutya |
| - harucha |
| - 春茶 |
| - rvcV2 |
| pipeline_tag: audio-to-audio |
| --- |
| |
| # Harutya (春茶) RVC V2 Model |
| Harutya/Harucha (春茶) channel link: https://www.youtube.com/@harutya |
|
|
| Need the dataset for some reason? Feel free to DM @jsj5 (Old: JSJ_15#8999) at Discord.\ |
| https://discord.com/users/572277409395638274 |
| |
| The dataset was about 40m long (raw 3hrs), it couldn't be short because Harutya\ |
| rarely sings in low pitches and high pitches so the dataset needed to be long. |
| |
| Q: Why are there voice cracks in the higher ends?\ |
| A: Harutya applies effects/dubs her voice on the parts which she sings in high & low pitches/tones\ |
| so I need to cut them out to maintain quality and consistency. |
| |
| Q: Why so breathy and noisy?\ |
| A: I think it is a special characteristic of her voice, and the reason why people usually listen to her covers.\ |
| Maybe try to reduce the noise on the input audio to keep maybe make it better?? |
| |
| |
| Checksums:\ |
| 51bac1a2154863a182d501969990b6f3 harutyaV2_e100_s100.pth\ |
| d275094bcee0e2dda668b363d35257c8 added_IVF3166_Flat_nprobe_1_harutyaV2_v2.index |
