Merge pull request #518 from alibaba-damo-academy/dev_wjm2

jmwang66

2023-05-16 6f7e27eb7c2d0a7649ec8f14d167c8da8e29f906

 funasr/fileio/sound_scp.py

@@ -2,10 +2,14 @@
from pathlib import Path
from typing import Union

import random
import numpy as np
import soundfile
import librosa
from typeguard import check_argument_types

import torch
import torchaudio

from funasr.fileio.read_text import read_2column_text

@@ -32,6 +36,7 @@
        always_2d: bool = False,
        normalize: bool = False,
        dest_sample_rate: int = 16000,
        speed_perturb: Union[list, tuple] = None,
    ):
        assert check_argument_types()
        self.fname = fname
@@ -40,6 +45,7 @@
        self.normalize = normalize
        self.data = read_2column_text(fname)
        self.dest_sample_rate = dest_sample_rate
        self.speed_perturb = speed_perturb

    def __getitem__(self, key):
        wav = self.data[key]
@@ -53,8 +59,17 @@
                wav, sr=self.dest_sample_rate, mono=self.always_2d, dtype=self.dtype
            )

        if self.speed_perturb is not None:
            speed = random.choice(self.speed_perturb)
            if speed != 1.0:
                array, _ = torchaudio.sox_effects.apply_effects_tensor(
                    torch.tensor(array).view(1, -1), rate,
                    [['speed', str(speed)], ['rate', str(rate)]])
                array = array.view(-1).numpy()

        if array.ndim==2:
            array=array.transpose((1, 0))

        return rate, array

    def get_path(self, key):