hehe

readme.md

1. Let's have everything as functional. Then wrap it as a layer.

2. All should work for both batch/non-batch ..with forcing

   • freq = height
   • time = width
   • channel = channel

3. Least dependency

• no librosa, no scipy.

functional.py

import torch



def STFT(src, n_fft, hop_length=None, window=None, **kwargs):
    """A wrapper for torch.stft with some preset parameters.
    Returned value keeps both real and imageinary parts.
    For STFT magnitude, see spectrogram

    """
    if hop_length is None:
        hop_length = n_fft // 4
    if window is None:
        window = torch.hann_window(n_fft)
    return torch.stft(src, n_fft, hop_length, window=window, **kwargs)


def spectrogram(src, n_fft, hop_length=None, window=None, power=2.0, **kwargs):
    """
    returns magnitude of spectrogram
    """
    stft = STFT(src, n_fft, hop_length=hop_length, window=window, **kwargs)
    return stft.pow(2).sum(-1).pow(power / 2.0)


def melspectrogram(src, n_fft, hop_length, window, sr, power=2.0, **kwargs):
    """
    returns melspectrogram
    """
    specgram = spectrogram(src, n_fft, hop_length, window, power=2.0, **kwargs)
    # TODO: mel basis matrix

def amplitude_to_db():
    """
    magnitude scaling
    """

def power_to_db():
    """
    magnitude scaling
    """
    
def pseudo_cqt():
    """computing pseudo-cqt
    """

# To inherit from torchaudio
def compose():

def scale():

def pad_trim():

def downmix():

def mu_law_encoding():

def mu_law_decoding():

layers.py

"""
wrap up all the functionals
"""