sakuraba_ema_gpt-sovits

Sakuraba Ema GPT-SoVITS

./reference_audio 含训练用音频和对应台词可用于推理。
./reference_audio contains audio clips for training and matching text, which can be utilized for inference.

如果想使用游戏内更多音频训练，all.list 包含所有游戏音频和台词的对应关系。音频文件请使用 AssetRipper 从游戏文件中提取。
If you wish to incorporate additional in-game audio for training, all.list provides a full mapping between all game audio files and their respective dialogues. Extract the audio files from the game using AssetRipper.

random_select.py 用于从 all.list 中随机选取台词条目，可指定角色和条数。
random_select.py can be used to randomly select dialogues from all.list, specifying the character and number of lines.

选取后的 list 可能包含无对应音频的台词。将 list 和音频文件放置在一起，用 validate.py 验证并从 list 中删除无音频的台词。
The list of selected dialogues may contain lines that do not match any audio file. place the list along side the audio files and use validate.py to validate and delete those lines from the list.