load_audio_text_image_video 存在bug

[shell-nlp]

FunASR/funasr/models/ct_transformer/model.py

Line 257 in c4ac64f

text = load_audio_text_image_video(data_in, data_type=kwargs.get("kwargs", "text"))[0]

这行代码可能存在bug,

load_audio_text_image_video 的返回值是个 torch 类型，

elif isinstance(data_or_path_or_list, np.ndarray):  # audio sample point
        data_or_path_or_list = torch.from_numpy(data_or_path_or_list)  # .squeeze()  # [n_samples,]

但是

FunASR/funasr/models/ct_transformer/model.py

Line 263 in c4ac64f

tokens = split_words(text, jieba_usr_dict=self.jieba_usr_dict)

这里 却对 torch tensor 类型进行 split_words

[shell-nlp]

这个bug 长期存在， 一直没有修复