| | |
|---|
| | | import numpy as np |
|---|
| | | import torch |
|---|
| | | import torchaudio |
|---|
| | | import soundfile |
|---|
| | | from torch.utils.data.dataset import IterableDataset |
|---|
| | | from typeguard import check_argument_types |
|---|
| | | import os.path |
|---|
| | |
|---|
| | | bytes = f.read() |
|---|
| | | return load_bytes(bytes) |
|---|
| | | |
|---|
| | | def load_wav(input): |
|---|
| | | try: |
|---|
| | | return torchaudio.load(input)[0].numpy() |
|---|
| | | except: |
|---|
| | | waveform, _ = soundfile.read(input, dtype='float32') |
|---|
| | | if waveform.ndim == 2: |
|---|
| | | waveform = waveform[:, 0] |
|---|
| | | return np.expand_dims(waveform, axis=0) |
|---|
| | | |
|---|
| | | DATA_TYPES = { |
|---|
| | | "sound": lambda x: torchaudio.load(x)[0].numpy(), |
|---|
| | | "sound": load_wav, |
|---|
| | | "pcm": load_pcm, |
|---|
| | | "kaldi_ark": load_kaldi, |
|---|
| | | "bytes": load_bytes, |
|---|
