update data2vec pretrain: dataset

jmwang66

2023-02-06 55b45487c75e683cafe63dba29e32be8243b6cc7

 funasr/datasets/large_datasets/utils/filter.py

@@ -6,13 +6,21 @@
           speech_length_max=15000,
           token_length_min=0,
           token_length_max=200):
    assert "speech" in data
    assert "text" in data
    assert "speech" in data or "text" in data

    if "sampling_rate" in data:
        speech_length = (data["speech"].shape[0] / data["sampling_rate"]) * 1000.
    if "speech" in data and "text" in data:
        if "sampling_rate" in data:
            speech_length = (data["speech"].shape[0] / data["sampling_rate"]) * 1000.
        else:
            speech_length = data["speech"].shape[0]
        num_tokens = len(data['text'])
        return speech_length_min < speech_length < speech_length_max and token_length_min < num_tokens < token_length_max
    elif "speech" in data:
        if "sampling_rate" in data:
            speech_length = (data["speech"].shape[0] / data["sampling_rate"]) * 1000.
        else:
            speech_length = data["speech"].shape[0]
        return speech_length_min < speech_length < speech_length_max
    else:
        speech_length = data["speech"].shape[0]
    num_tokens = len(data['text'])

    return speech_length_min < speech_length < speech_length_max and token_length_min < num_tokens < token_length_max
        num_tokens = len(data['text'])
        return token_length_min < num_tokens < token_length_max