Last active
February 19, 2024 08:30
-
-
Save mthrok/01f89d9bc27a7fe618bf5e8ef71b44ba to your computer and use it in GitHub Desktop.
Compare spectrograms of torchaudio and librosa
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
import torch | |
import torchaudio | |
import librosa | |
import matplotlib.pyplot as plt | |
from torchaudio.functional import create_fb_matrix | |
n_fft = 2048 | |
n_mels = 128 | |
sample_rate = 6000 | |
torchaudio_mel = create_fb_matrix( | |
int(n_fft // 2 + 1), | |
n_mels=n_mels, | |
f_min=0., | |
f_max=sample_rate/2., | |
sample_rate=sample_rate, | |
norm='slaney' | |
) | |
librosa_mel = librosa.filters.mel( | |
sample_rate, | |
n_fft, | |
n_mels=n_mels, | |
fmin=0., | |
fmax=sample_rate/2., | |
norm='slaney', | |
htk=True, | |
).T | |
mse = ((torchaudio_mel - librosa_mel) ** 2).mean() | |
print(f'MSE:\t{mse}') | |
fig, axs = plt.subplots(1, 2, figsize=(10, 10)) | |
fig.suptitle('mel-filter bank') | |
axs[0].set_title('torchaudio[slaney]') | |
axs[0].imshow(torchaudio_mel, aspect='auto') | |
axs[0].set_ylabel('frequency bin') | |
axs[0].set_xlabel('mel bin') | |
axs[1].set_title('librosa[htk + slaney]') | |
axs[1].imshow(librosa_mel, aspect='auto') | |
axs[0].set_ylabel('frequency bin') | |
axs[1].set_xlabel('mel bin') | |
plt.show() |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
import librosa | |
import matplotlib.pyplot as plt | |
import torch | |
from torchaudio.transforms import MelSpectrogram | |
n_fft = 2048 | |
win_len = None | |
hop_len = 512 | |
n_mels = 128 | |
sample_rate = 6000 | |
path = 'test/torchaudio_unittest/assets/steam-train-whistle-daniel_simon.wav' | |
waveform, sample_rate = librosa.load(path, sr=sample_rate) | |
waveform = torch.Tensor(waveform) | |
torchaudio_melspec = MelSpectrogram( | |
sample_rate=sample_rate, | |
n_fft=n_fft, | |
win_length=win_len, | |
hop_length=hop_len, | |
center=True, | |
pad_mode="reflect", | |
power=2.0, | |
norm='slaney', | |
onesided=True, | |
n_mels=n_mels, | |
)(waveform) | |
librosa_melspec = librosa.feature.melspectrogram( | |
waveform.numpy(), | |
sr=sample_rate, | |
n_fft=n_fft, | |
hop_length=hop_len, | |
win_length=win_len, | |
center=True, | |
pad_mode="reflect", | |
power=2.0, | |
n_mels=n_mels, | |
norm='slaney', | |
htk=True, | |
) | |
mse = ((torchaudio_melspec - librosa_melspec) ** 2).mean() | |
print(f'MSE:\t{mse}') | |
fig, axs = plt.subplots(1, 2, figsize=(20, 5)) | |
fig.suptitle('Mel Spectrogram') | |
axs[0].set_title('torchaudio') | |
axs[0].set_ylabel('mel bin') | |
axs[0].set_xlabel('frame') | |
axs[0].imshow(librosa.power_to_db(torchaudio_melspec), aspect='auto') | |
axs[1].set_title('librosa') | |
axs[1].set_ylabel('mel bin') | |
axs[1].set_xlabel('frame') | |
axs[1].imshow(librosa.power_to_db(librosa_melspec), aspect='auto') | |
plt.show() |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
import librosa.core.spectrum | |
import matplotlib.pyplot as plt | |
import torch | |
from torchaudio.transforms import Spectrogram | |
n_fft = 2048 | |
win_len = None | |
hop_len = 512 | |
sample_rate = 6000 | |
path = 'test/torchaudio_unittest/assets/steam-train-whistle-daniel_simon.wav' | |
waveform, _ = librosa.load(path, sr=sample_rate) | |
waveform = torch.Tensor(waveform) | |
torchaudio_spec = Spectrogram( | |
n_fft=n_fft, | |
win_length=win_len, | |
hop_length=hop_len, | |
center=True, | |
pad_mode="reflect", | |
power=2.0, | |
)(waveform) | |
librosa_spec, _ = librosa.core.spectrum._spectrogram( | |
waveform.numpy(), | |
n_fft=n_fft, | |
hop_length=hop_len, | |
win_length=win_len, | |
center=True, | |
pad_mode="reflect", | |
power=2.0, | |
) | |
mse = ((torchaudio_spec - librosa_spec) ** 2).mean() | |
print(f'MSE:\t{mse}') | |
fig, axs = plt.subplots(1, 2, figsize=(20, 5)) | |
fig.suptitle('mel-Kernel') | |
axs[0].set_title('torchaudio') | |
axs[0].set_ylabel('mel bin') | |
axs[0].set_xlabel('frame') | |
axs[0].imshow(librosa.power_to_db(torchaudio_spec), aspect='auto') | |
axs[1].set_title('librosa') | |
axs[1].set_ylabel('mel bin') | |
axs[1].set_xlabel('frame') | |
axs[1].imshow(librosa.power_to_db(librosa_spec), aspect='auto') | |
plt.show() |
Not among the set of compatible parameters. Please refer to https://pytorch.org/audio/0.10.0/tutorials/audio_feature_extractions_tutorial.html#mel-filter-bank for a better explanation.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Are there any noticeable differences you've observed?