nnAudio.features.mel.MelSpectrogram

class nnAudio.features.mel.MelSpectrogram(sr=22050, n_fft=2048, win_length=None, n_mels=128, hop_length=512, window='hann', center=True, pad_mode='reflect', power=2.0, htk=False, fmin=0.0, fmax=None, norm=1, trainable_mel=False, trainable_STFT=False, verbose=True, **kwargs)

Bases: torch.nn.modules.module.Module

This function is to calculate the Melspectrogram of the input signal. Input signal should be in either of the following shapes.

  1. (len_audio)

  2. (num_audio, len_audio)

  3. (num_audio, 1, len_audio)

The correct shape will be inferred automatically if the input follows these 3 shapes. Most of the arguments follow the convention from librosa. This class inherits from nn.Module, therefore, the usage is same as nn.Module.

Parameters
  • sr (int) – The sampling rate for the input audio. It is used to calculate the correct fmin and fmax. Setting the correct sampling rate is very important for calculating the correct frequency.

  • n_fft (int) – The window size for the STFT. Default value is 2048

  • win_length (int) – the size of window frame and STFT filter. Default: None (treated as equal to n_fft)

  • n_mels (int) – The number of Mel filter banks. The filter banks maps the n_fft to mel bins. Default value is 128.

  • hop_length (int) – The hop (or stride) size. Default value is 512.

  • window (str) – The windowing function for STFT. It uses scipy.signal.get_window, please refer to scipy documentation for possible windowing functions. The default value is ‘hann’.

  • center (bool) – Putting the STFT keneral at the center of the time-step or not. If False, the time index is the beginning of the STFT kernel, if True, the time index is the center of the STFT kernel. Default value if True.

  • pad_mode (str) – The padding method. Default value is ‘reflect’.

  • htk (bool) – When False is used, the Mel scale is quasi-logarithmic. When True is used, the Mel scale is logarithmic. The default value is False.

  • fmin (int) – The starting frequency for the lowest Mel filter bank.

  • fmax (int) – The ending frequency for the highest Mel filter bank.

  • norm – if 1, divide the triangular mel weights by the width of the mel band (area normalization, AKA ‘slaney’ default in librosa). Otherwise, leave all the triangles aiming for a peak value of 1.0

  • trainable_mel (bool) – Determine if the Mel filter banks are trainable or not. If True, the gradients for Mel filter banks will also be calculated and the Mel filter banks will be updated during model training. Default value is False.

  • trainable_STFT (bool) – Determine if the STFT kenrels are trainable or not. If True, the gradients for STFT kernels will also be caluclated and the STFT kernels will be updated during model training. Default value is False.

  • verbose (bool) – If True, it shows layer information. If False, it suppresses all prints.

Returns

spectrogram – It returns a tensor of spectrograms. shape = (num_samples, freq_bins,time_steps).

Return type

torch.tensor

Examples

>>> spec_layer = Spectrogram.MelSpectrogram()
>>> specs = spec_layer(x)

Methods

__init__

Initializes internal Module state, shared by both nn.Module and ScriptModule.

extra_repr

Set the extra representation of the module

forward

Convert a batch of waveforms to Mel spectrograms.

extra_repr()str

Set the extra representation of the module

To print customized extra information, you should reimplement this method in your own modules. Both single-line and multi-line strings are acceptable.

forward(x)

Convert a batch of waveforms to Mel spectrograms.

Parameters

x (torch tensor) –

Input signal should be in either of the following shapes.

  1. (len_audio)

  2. (num_audio, len_audio)

3. (num_audio, 1, len_audio) It will be automatically broadcast to the right shape