Emotional issues in voice-over ~
#3
by sunnyboxs - opened
The LoRA performs quite well; however, in my comparative tests—for instance, when using a prompt like (in Chinese):
男人的画外音说:"测试角色音色,这个效果真不错",
—the voice-over generated by DMAD sounds quite strange.
It lacks emotion, feels mechanical, and carries a very distinct "text-to-speech" (TTS) quality.
I also tested other LoRAs; using the same prompts, the emotional nuances differed significantly.
Could the use of TTS during training have caused the emotional expressiveness of the voice-over to weaken?
Thanks~
It might have been an issue with the model; changing the random seed resolved it.
sunnyboxs changed discussion status to closed