Paddleocr Finetuning advice please

Hello, 
i wanna finetune the en_PP-OCRv4_rec_infer model. my question is, am I able to change the possible Output symbols by simply just changing character_dict_path to my custom txt file containing less symbols. My goal is even if the model recognizes a symbol like "%" I dont want it to be printed.