def __init__(
self,
mixture_path: str,
audio_sample_rate: int,
sample_length: int,
stft_n_fft: int,
stft_win_length: int,
stft_hop: int,
device: torch.device = torch.device("cpu")) -> None:
self.sample_length: int = sample_length
self.audio_sample_rate: int = audio_sample_rate
self.stft_n_fft: int = stft_n_fft
self.stft_win_length = stft_win_length
self.stft_hop: int = stft_hop
waveform_length = Audio.calc_waveform_length(mixture_path,
audio_sample_rate)
self.samples_count = Audio.calc_stft_length(waveform_length, stft_hop)
self.waveform = Audio.load(mixture_path,
sample_rate=audio_sample_rate,
device=device)
self.amplitude, self.phase = Audio.compute_stft(
self.waveform, self.stft_n_fft, self.stft_win_length,
self.stft_hop)
def __init__(
self,
data_frame: pandas.DataFrame,
audio_sample_rate: int,
sample_length: int,
stft_n_fft: int,
stft_win_length: int,
stft_hop: int,
shuffle: bool = False,
seed: Any = None,
device: torch.device = torch.device("cpu")) -> None:
self.data = pandas.DataFrame()
self.audio_sample_rate: int = audio_sample_rate
self.sample_length: int = sample_length
self.stft_n_fft: int = stft_n_fft
self.stft_win_length: int = stft_win_length
self.stft_hop: int = stft_hop
self.shuffle: bool = shuffle
self.seed: Any = seed if (seed is not None) else random.random()
self.device = device
for i, row in data_frame.iterrows():
new_row = dict()
for col in ["mixture", "vocals", "accompaniment"]:
path = pathlib.Path(row[col]).absolute()
if not path.exists() or path.is_dir():
raise FileNotFoundError(path)
new_row[col] = str(path)
waveform_length = Audio.calc_waveform_length(
row["mixture"], self.audio_sample_rate)
stft_length = Audio.calc_stft_length(waveform_length,
self.stft_hop)
new_row["n_frames"] = stft_length
self.data = self.data.append(new_row, ignore_index=True)
