V2ProPlus Fine-tune & LoRA Question

hello,

(1) For a model v2proplus, is it possible to implement macro-level(each sample) tone control and fine-grained special tokens (like (laugh)) through fine-tuning alone?

Or would adding these tokens and emotional embeddings necessitate a full phase from scratch?

ex)

text,tone,wav "hello (laugh) i'm elsa","","00001.wav"

(2) Does GPTSoVITS supports LoRA(adapter) request ?

(3) When fine-tuning for 3 specific speakers, which approach is best for performance and efficiency? a. Fine-tuning a single model using a combined dataset of all 3 speakers at once. b. Fine-tuning a single base model on Speaker A, then B, then C in a series. c. Fine-tuning a each model for each speaker Which method is most effective in terms of speaker identity and server resource optimization ??

Thank you !