We use cookies on this website to distinguish you from other users. We use this data to improve our content experience and for targeted advertising. By continuing to use this website you consent to our use of cookies. For more information, please see our Cookie Policy.

spafe.features.lfcc#

Description : Linear Frequency Cepstral Coefﬁcients (LFCCs) extraction algorithm implementation.
Copyright (c) 2019-2024 Ayoub Malek. This source code is licensed under the terms of the BSD 3-Clause License. For a copy, see <https://github.com/SuperKogito/spafe/blob/master/LICENSE>.

spafe.features.lfcc.linear_spectrogram(sig: ndarray, fs: int = 16000, pre_emph: bool = True, pre_emph_coeff: float = 0.97, window: Optional[SlidingWindow] = None, nfilts: int = 24, nfft: int = 512, low_freq: float = 0, high_freq: Optional[float] = None, scale: typing_extensions.Literal[ascendant, descendant, constant] = 'constant', fbanks: Optional[ndarray] = None) → ndarray[source]#

Compute the mel scale spectrogram.

Parameters

sig (numpy.ndarray) – a mono audio signal (Nx1) from which to compute features.
fs (int) – the sampling frequency of the signal we are working with. (Default is 16000).
pre_emph (bool) – apply pre-emphasis if 1. (Default is 1).
pre_emph_coeff (float) – pre-emphasis filter coefficient. (Default is 0.97).
window (SlidingWindow) – sliding window object. (Default is None).
nfilts (int) – the number of filters in the filter bank. (Default is 40).
nfft (int) – number of FFT points. (Default is 512).
low_freq (float) – lowest band edge of mel filters (Hz). (Default is 0).
high_freq (float) – highest band edge of mel filters (Hz). (Default is samplerate / 2).
scale (str) – monotonicity behavior of the filter banks. (Default is “constant”).
fbanks (numpy.ndarray) – filter bank matrix. (Default is None).

Returns

features - the LFFC features: num_frames x num_ceps

Return type

(numpy.ndarray)

Tip

scale : can take the following options [“constant”, “ascendant”, “descendant”].

Note

../_images/linear_spectrogram.png — Architecture of linear scale spectrogram extraction algorithm.#

Examples

from spafe.features.lfcc import linear_spectrogram
from spafe.utils.vis import show_spectrogram
from spafe.utils.preprocessing import SlidingWindow
from scipy.io.wavfile import read

# read audio
fpath = "../../../tests/data/test.wav"
fs, sig = read(fpath)

# compute spectrogram
lSpec, lfreqs = linear_spectrogram(sig,
                                   fs=fs,
                                   pre_emph=0,
                                   pre_emph_coeff=0.97,
                                   window=SlidingWindow(0.03, 0.015, "hamming"),
                                   nfilts=128,
                                   nfft=2048,
                                   low_freq=0,
                                   high_freq=fs/2)

# visualize spectrogram
show_spectrogram(lSpec.T,
                 fs,
                 xmin=0,
                 xmax=len(sig)/fs,
                 ymin=0,
                 ymax=(fs/2)/1000,
                 dbf=80.0,
                 xlabel="Time (s)",
                 ylabel="Frequency (kHz)",
                 title="Linear spectrogram (dB)",
                 cmap="jet")

spafe.features.lfcc.lfcc(sig, fs: int = 16000, num_ceps=13, pre_emph: bool = True, pre_emph_coeff: float = 0.97, window: Optional[SlidingWindow] = None, nfilts: int = 24, nfft: int = 512, low_freq: float = 0, high_freq: Optional[float] = None, scale: typing_extensions.Literal[ascendant, descendant, constant] = 'constant', dct_type=2, use_energy=False, lifter: Optional[int] = None, normalize: Optional[typing_extensions.Literal[mvn, ms, vn, mn]] = None, fbanks: Optional[ndarray] = None)[source]#

Compute the linear-frequency cepstral coefﬁcients (GFCC features) from an audio signal as described in [Sehili].

Parameters

sig (numpy.ndarray) – a mono audio signal (Nx1) from which to compute features.
fs (int) – the sampling frequency of the signal we are working with. (Default is 16000).
num_ceps (int) – number of cepstra to return. (Default is 13).
pre_emph (bool) – apply pre-emphasis if 1. (Default is 1).
pre_emph_coeff (float) – pre-emphasis filter coefficient. (Default is 0.97).
window (SlidingWindow) – sliding window object. (Default is None).
nfilts (int) – the number of filters in the filter bank. (Default is 40).
nfft (int) – number of FFT points. (Default is 512).
low_freq (float) – lowest band edge of mel filters (Hz). (Default is 0).
high_freq (float) – highest band edge of mel filters (Hz). (Default is samplerate / 2).
scale (str) – monotonicity behavior of the filter banks. (Default is “constant”).
dct_type (int) – type of DCT used. (Default is 2).
use_energy (bool) – overwrite C0 with true log energy. (Default is 0).
lifter (int) – apply liftering if specified. (Default is None).
normalize (str) – apply normalization if approach provided. (Default is None).
fbanks (numpy.ndarray) – filter bank matrix. (Default is None).

Returns

2d array of LFCC features (num_frames x num_ceps).

Return type

(numpy.ndarray)

Tip

scale : can take the following options [“constant”, “ascendant”, “descendant”].
dct : can take the following options [1, 2, 3, 4].
normalize : can take the following options [“mvn”, “ms”, “vn”, “mn”].

Note

../_images/lfccs.png — Architecture of linear frequency cepstral coefﬁcients extraction algorithm.#

Examples

from scipy.io.wavfile import read
from spafe.features.lfcc import lfcc
from spafe.utils.preprocessing import SlidingWindow
from spafe.utils.vis import show_features

# read audio
fpath = "../../../tests/data/test.wav"
fs, sig = read(fpath)

# compute lfccs
lfccs  = lfcc(sig,
              fs=fs,
              pre_emph=1,
              pre_emph_coeff=0.97,
              window=SlidingWindow(0.03, 0.015, "hamming"),
              nfilts=128,
              nfft=2048,
              low_freq=0,
              high_freq=8000,
              normalize="mvn")

# visualize features
show_features(lfccs, "Linear Frequency Cepstral Coefﬁcients", "LFCC Index", "Frame Index")

References

Sehili: : Sehili, A. & Istrate, B. & Boudy, J. (2022). Primary Investigation of Sound Recognition for a domotic application using Support Vector.

spafe.features.gfcc

spafe.features.lpc