Merge pull request #687 from alibaba-damo-academy/dev_lhn

hnluo

2023-06-29 c2dee5e3c29eba79e591d9e9caebaef15ea4e56b

 funasr/datasets/small_datasets/dataset.py

@@ -9,8 +9,7 @@
from typing import Collection
from typing import Dict
from typing import Mapping
from typing import Tuple
from typing import Union
from typing import Union, List, Tuple

import kaldiio
import numpy as np
@@ -110,7 +109,7 @@
            float_dtype: str = "float32",
            int_dtype: str = "long",
            dest_sample_rate: int = 16000,
            speed_perturb: tuple = None,
            speed_perturb: Union[list, tuple] = None,
            mode: str = "train",
    ):
        assert check_argument_types()
@@ -144,7 +143,7 @@

    def _build_loader(
            self, path: str, loader_type: str
    ) -> Mapping[str, Union[np.ndarray, torch.Tensor, str, numbers.Number]]:
    ) -> Mapping[str, Union[np.ndarray, torch.Tensor, str, List[int], numbers.Number]]:
        """Helper function to instantiate Loader.

        Args:
@@ -174,6 +173,19 @@
                        raise RuntimeError(f"{k} is duplicated ({path}:{linenum})")
                    text_loader[k] = v
            return text_loader
        elif loader_type == "text_int":
            text_int_loader = {}
            with open(path, "r", encoding="utf-8") as f:
                for linenum, line in enumerate(f, 1):
                    sps = line.rstrip().split(maxsplit=1)
                    if len(sps) == 1:
                        k, v = sps[0], ""
                    else:
                        k, v = sps
                    if k in text_int_loader:
                        raise RuntimeError(f"{k} is duplicated ({path}:{linenum})")
                    text_int_loader[k] = [int(i) for i in v.split()]
            return text_int_loader
        else:
            raise RuntimeError(f"Not supported: loader_type={loader_type}")