funasr sdk

游雁

2023-07-05 4e2fe544ae37174a3e09dfcdbbdae5abfe711e53

 funasr/datasets/large_datasets/dataset.py

@@ -128,7 +128,9 @@
                        try:
                            waveform, sampling_rate = torchaudio.load(path)
                        except:
                            waveform, sampling_rate = soundfile.read(path)
                            waveform, sampling_rate = soundfile.read(path, dtype='float32')
                            if waveform.ndim == 2:
                                waveform = waveform[:, 0]
                            waveform = np.expand_dims(waveform, axis=0)
                            waveform = torch.tensor(waveform)
                        if self.frontend_conf is not None:
@@ -200,14 +202,7 @@
    data_types = conf.get("data_types", "kaldi_ark,text")

    pre_hwfile = conf.get("pre_hwlist", None)
    pre_prob = conf.get("pre_prob", 0)  # unused yet

    hw_config = {"sample_rate": conf.get("sample_rate", 0.6),
                 "double_rate": conf.get("double_rate", 0.1),
                 "hotword_min_length": conf.get("hotword_min_length", 2),
                 "hotword_max_length": conf.get("hotword_max_length", 8),
                 "pre_prob": conf.get("pre_prob", 0.0)}

    # pre_prob = conf.get("pre_prob", 0)  # unused yet
    if pre_hwfile is not None:
        pre_hwlist = []
        with open(pre_hwfile, 'r') as fin:
@@ -216,6 +211,15 @@
    else:
        pre_hwlist = None

    hw_config = {"sample_rate": conf.get("sample_rate", 0.6),
                 "double_rate": conf.get("double_rate", 0.1),
                 "hotword_min_length": conf.get("hotword_min_length", 2),
                 "hotword_max_length": conf.get("hotword_max_length", 8),
                 "pre_prob": conf.get("pre_prob", 0.0),
                 "pre_hwlist": pre_hwlist}

    

    dataset = AudioDataset(scp_lists, 
                           data_names, 
                           data_types,