update repo

嘉渊

2023-05-16 12dd848db2cfd0e2ae6f32cfb1a5aecdf0f77365

 funasr/datasets/small_datasets/dataset.py

@@ -9,6 +9,7 @@
from typing import Collection
from typing import Dict
from typing import Mapping
from typing import Optional
from typing import Tuple
from typing import Union

@@ -110,6 +111,8 @@
            float_dtype: str = "float32",
            int_dtype: str = "long",
            dest_sample_rate: int = 16000,
            speed_perturb: Union[list, tuple] = None,
            mode: str = "train",
    ):
        assert check_argument_types()
        if len(path_name_type_list) == 0:
@@ -123,6 +126,10 @@
        self.float_dtype = float_dtype
        self.int_dtype = int_dtype
        self.dest_sample_rate = dest_sample_rate
        self.speed_perturb = speed_perturb
        self.mode = mode
        if self.speed_perturb is not None:
            logging.info("Using speed_perturb: {}".format(speed_perturb))

        self.loader_dict = {}
        self.debug_info = {}
@@ -146,13 +153,15 @@
            loader_type:  loader_type. sound, npy, text, etc
        """
        if loader_type == "sound":
            loader = SoundScpReader(path, self.dest_sample_rate, normalize=True, always_2d=False)
            speed_perturb = self.speed_perturb if self.mode == "train" else None
            loader = SoundScpReader(path, self.dest_sample_rate, normalize=True, always_2d=False,
                                    speed_perturb=speed_perturb)
            return AdapterForSoundScpReader(loader, self.float_dtype)
        elif loader_type == "kaldi_ark":
            loader = kaldiio.load_scp(path)
            return AdapterForSoundScpReader(loader, self.float_dtype)
        elif loader_type == "npy":
            return NpyScpReader()
            return NpyScpReader(path)
        elif loader_type == "text":
            text_loader = {}
            with open(path, "r", encoding="utf-8") as f: