Merge pull request #408 from alibaba-damo-academy/hnluo-patch-1

zhifu gao

2023-04-24 331d57253ae25dd42c8e14930dee30cd8d2affa6

 funasr/datasets/large_datasets/utils/tokenize.py

@@ -19,6 +19,7 @@
def seg_tokenize(txt, seg_dict):
    out_txt = ""
    for word in txt:
        word = word.lower()
        if word in seg_dict:
            out_txt += seg_dict[word] + " "
        else:
@@ -28,23 +29,26 @@
def tokenize(data,
             vocab=None,
             seg_dict=None,
             punc_dict=None):
             punc_dict=None,
             bpe_tokenizer=None):
    assert "text" in data
    assert isinstance(vocab, dict)
    text = data["text"]
    token = []
    vad = -2

    if bpe_tokenizer is not None:
        text = bpe_tokenizer.text2tokens("".join(text))

    if seg_dict is not None:
        assert isinstance(seg_dict, dict)
        txt = forward_segment("".join(text).lower(), seg_dict)
        text = seg_tokenize(txt, seg_dict)
        text = seg_tokenize(text, seg_dict)

    length = len(text)
    for i in range(length):
        x = text[i]
        if i == length-1 and "punc" in data and text[i].startswith("vad:"):
            vad = x[-1][4:]
        if i == length-1 and "punc" in data and x.startswith("vad:"):
            vad = x[4:]
            if len(vad) == 0:
                vad = -1
            else: