Mel Spectrogram

Create MelSpectrogram for a raw audio signal. This is a composition of Spectrogram and MelScale.

transform_mel_spectrogram(
  sample_rate = 16000,
  n_fft = 400,
  win_length = NULL,
  hop_length = NULL,
  f_min = 0,
  f_max = NULL,
  pad = 0,
  n_mels = 128,
  window_fn = torch::torch_hann_window,
  power = 2,
  normalized = FALSE,
  ...
)

Arguments

sample_rate

(int, optional): Sample rate of audio signal. (Default: 16000)

n_fft

(int, optional): Size of FFT, creates n_fft // 2 + 1 bins. (Default: 400)

win_length

(int or NULL, optional): Window size. (Default: n_fft)

hop_length

(int or NULL, optional): Length of hop between STFT windows. (Default: win_length // 2)

f_min

(float, optional): Minimum frequency. (Default: 0.)

f_max

(float or NULL, optional): Maximum frequency. (Default: NULL)

pad

(int, optional): Two sided padding of signal. (Default: 0)

n_mels

(int, optional): Number of mel filterbanks. (Default: 128)

window_fn

(function, optional): A function to create a window tensor that is applied/multiplied to each frame/window. (Default: torch_hann_window)

power

(float, optional): Power of the norm. (Default: to 2.0)

normalized

(logical): Whether to normalize by magnitude after stft (Default: FALSE)

...

(optional): Arguments for window function.

Value

tensor: Mel frequency spectrogram of size (..., n_mels, time).

Details

forward param: waveform (Tensor): Tensor of audio of dimension (..., time).

Sources

• https://gist.github.com/kastnerkyle/179d6e9a88202ab0a2fe

• https://timsainb.github.io/spectrograms-mfccs-and-inversion-in-python.html

• https://haythamfayek.com/2016/04/21/speech-processing-for-machine-learning.html

Examples

#'   Example
if (FALSE) {

if(torch::torch_is_installed()) {
mp3_path <- system.file("sample_audio_1.mp3", package = "torchaudio")
sample_mp3 <- transform_to_tensor(tuneR_loader(mp3_path))
# (channel, n_mels, time)
mel_specgram <- transform_mel_spectrogram(sample_rate = sample_mp3[[2]])(sample_mp3[[1]])
}
}