modulation.torch_utilities.torch_transforms

Classes

InversePreEmphasis([coef])

Implement Inverse Pre-emphasis by using RNN to boost up inference speed.

InverseSpectrogram(n_fft, win_length, hop_length)

Convert from magphase to complex and perform istft

LogTransform([fill_value])

Transform for taking logarithm of mel spectrograms (or anything else) :param fill_value: value to substitute non-positive numbers with before applying log

MelSpectrogram([clip_min_value])

torchaudio MelSpectrogram wrapper for audiomentations's Compose

PreEmphasis([coef])

description

Spectrogram(n_fft, win_length, hop_length)

Apply stft and magphase transformations

Squeeze()

Transform to squeeze mono channel waveform

ToMono()

Convert stereo signal to mono

ToNumpy()

Transform to make numpy array

ToTorch()

Transform to make torch.tensor