Dev kws (#2105)

zhifu gao

2024-09-25 2196844d1d6e5b8732c95896bb46f0eacdd9cf9d

 funasr/datasets/audio_datasets/datasets.py

@@ -1,6 +1,7 @@
import torch
import random


from funasr.register import tables
from funasr.utils.load_utils import extract_fbank, load_audio_text_image_video

@@ -17,6 +18,7 @@
        index_ds: str = None,
        frontend=None,
        tokenizer=None,
        is_training: bool = True,
        int_pad_value: int = -1,
        float_pad_value: float = 0.0,
        **kwargs,
@@ -24,6 +26,11 @@
        super().__init__()
        index_ds_class = tables.index_ds_classes.get(index_ds)
        self.index_ds = index_ds_class(path, **kwargs)

        self.preprocessor_speech = None
        self.preprocessor_text = None

        if is_training:
        preprocessor_speech = kwargs.get("preprocessor_speech", None)
        if preprocessor_speech:
            preprocessor_speech_class = tables.preprocessor_classes.get(preprocessor_speech)
@@ -64,6 +71,7 @@
        data_src = load_audio_text_image_video(source, fs=self.fs)
        if self.preprocessor_speech:
            data_src = self.preprocessor_speech(data_src, fs=self.fs)

        speech, speech_lengths = extract_fbank(
            data_src, data_type=self.data_type, frontend=self.frontend, is_final=True
        )  # speech: [b, T, d]
@@ -71,6 +79,7 @@
        target = item["target"]
        if self.preprocessor_text:
            target = self.preprocessor_text(target)

        if self.tokenizer:
            ids = self.tokenizer.encode(target)
            text = torch.tensor(ids, dtype=torch.int64)