initially i was going for 475e but then it sounded awful and the 350e version sounded a bit more stable so i went with that.
it's also a test to see how the model trained on klm pretrain would handle lossy audio as dataset
this character's featured in blue archive.
trained on 11 minutes of audio data from both jp and global versions since both of them have different lines. (eq + spectral de-noise)
CV. Tanezaki Atsumi
pitch extraction: mangio-crepe / (applio) crepe - 64 hops
steps: 14k
batch size: 7
pretrain: klm 4 hfg / 40k
hf link: https://huggingface.co/sxndypz/rvc-v2-models/resolve/main/seia.zip
Tags: No tags available
Download Link: https://huggingface.co/sxndypz/rvc-v2-models/resolve/main/seia.zip