spafe.features.spfeats#
Description : Spectral and frequency stats and features extraction algorithms implementation.
Copyright (c) 2019-2023 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>.
- class spafe.features.spfeats.SpectralFeats(_typename, _fields=None, /, **kwargs)[source]#
Bases:
dict
- spafe.features.spfeats.spectral_centroid(fs: int, spectrum: ndarray, i: int = 1) float[source]#
Compute the spectral centroid (which is the barycenter of the spectrum) as described in [Peeters].
- Parameters
fs (int) β input signal sampling rate.
spectrum (numpy.ndarray) β signal spectrum.
i (int) β centroid order. (Default is 1).
- Returns
spectral centroid.
- Return type
(float)
Note
\[\begin{split}\mu_{i} &= \frac{\sum_{n=0}^{N}f_{k}^{i}*a_{k}}{\sum_{n=0}^{N}a_{k}}\\ S_{centroid} &= \mu_{i}\\\end{split}\]
- spafe.features.spfeats.spectral_skewness(sig: ndarray, fs: int, spectrum: ndarray)[source]#
Compute the spectral skewness (which is a measure of the asymmetry of a distribution around its mean) as described in [Peeters].
- Parameters
sig (numpy.ndarray) β input mono signal.
fs (int) β input signal sampling rate.
spectrum (numpy.ndarray) β signal spectrum.
- Returns
spectral skewness.
- Return type
(float)
Note
\[\begin{split}S_{skewness} = \mu_{3} &= \frac{\sum_{n=0}^{N}(f_{k} - \mu_{1})^{3} . a_k}{\mu_{2}^{3} . \sum_{n=0}^{N}a_{k}}\\\end{split}\]
- spafe.features.spfeats.spectral_kurtosis(sig: ndarray, fs: int, spectrum: ndarray) float[source]#
Compute the spectral kurtosis (which is a measure of the flatness of a distribution around its mean) as described in [Peeters].
- Parameters
sig (numpy.ndarray) β input mono signal.
fs (int) β input signal sampling rate.
spectrum (numpy.ndarray) β signal spectrum.
- Returns
spectral kurtosis.
- Return type
(float)
Note
\[\begin{split}S_{kurtosis} = \mu_{4} &= \frac{\sum_{n=0}^{N}(f_{k} - \mu_{1})^{4} . a_k}{\mu_{2}^{4} . \sum_{n=0}^{N}a_{k}}\\\end{split}\]
- spafe.features.spfeats.spectral_entropy(spectrum: ndarray) float[source]#
Compute the spectral entropy as described in [Misra].
- Parameters
spectrum (numpy.ndarray) β signal spectrum.
- Returns
spectral entropy.
- Return type
(float)
Note
\[\begin{split}S_{entropy} &= \frac{\sum_{n=0}^{N}(f_{k} - \mu_{1})^{3} . a_k}{\mu_{2}^{3} . \sum_{n=0}^{N}a_{k}}\\\end{split}\]References
- Misra
: Misra, H., S. Ikbal, H. Bourlard, and H. Hermansky. βSpectral Entropy Based Feature for Robust ASR.β 2004 IEEE International Conference on Acoustics, Speech, and Signal Processing.
- spafe.features.spfeats.spectral_spread(sig: ndarray, fs: int, spectrum: ndarray) float[source]#
Compute the spectral spread (basically a variance of the spectrum around the spectral centroid) as described in [Peeters].
- Parameters
sig (numpy.ndarray) β input mono signal.
fs (int) β input signal sampling rate.
spectrum (numpy.ndarray) β signal spectrum.
- Returns
spectral spread.
- Return type
(float)
Note
\[S_{spread} = \sqrt{\sum{0}{N}\frac{(f_k - \mu_{1}) . s_k}{\sum{0}{N}a_k}}\]
- spafe.features.spfeats.spectral_flatness(spectrum: ndarray) float[source]#
Compute spectral flatness.
Args:
# spectrum (numpy.ndarray) : signal spectrum.
- Returns:
(float) : spectral flatness value.
- Note:
- \[S_{flatness} = \frac{exp(\frac{1}{N}\sum_{k}log(a_{k}))}{\frac{1}{N}\sum_{k}a_{k}}\]
- spafe.features.spfeats.spectral_rolloff(spectrum: ndarray, k: float = 0.85) float[source]#
Compute the spectral roll-off point which measures the bandwidth of the audio signal by determining the frequency bin under which a specified k percentage of the total spectral energy is contained below. see [Scheirer].
- Parameters
spectrum (numpy.ndarray) β signal spectrum.
k (float) β constant. (Default is 0.85).
- Returns
spectral rolloff point.
- Return type
(float)
- spafe.features.spfeats.spectral_flux(spectrum: ndarray, p: int = 2) float[source]#
Compute the spectral flux, which measures how quickly the power spectrum of a signal is changing. This implementation computes the spectral flux using the L2-norm per default i.e. the square root of the sum of absolute differences squared [Scheirer].
- Parameters
spectrum (numpy.ndarray) β signal spectrum.
p (int) β norm type. (Default is 2).
- Returns
spectral flux.
- Return type
(float)
Note
\[S_{flux} = (\sum_{k}(|a_{k}(t) - a_{k}(t-1))^2}{\sqrt{\sum_{k}a_{k}(t-1)|^p)^{\frac{1}{p}}\]References
- spafe.features.spfeats.extract_feats(sig: ndarray, fs: int, nfft: int = 512) SpectralFeats[source]#
Compute various spectral features [Peeters].
- Parameters
sig (numpy.ndarray) β input mono signal.
fs (int) β input signal sampling rate.
nfft (int) β number of FFT points. (Default is 512).
- Returns
a dictionary including various frequency and spectral features (see Notes).
- Return type
(dict)
Note
The resulting dictionary includes the following elements:
- spectral features
spectral_centroid: spectral centroid.spectral_skewness: frequencies skewness.spectral_kurtosis: frequencies kurtosis.spectral_entropy: spectral entropy.spectral_spread: spectral spread.spectral_flatness: spectral flatness.spectral_rolloff: spectral rolloff.spectral_flux: spectral flux.spectral_mean: spectral mean.spectral_rms: spectral root mean square.spectral_std: spectral standard deviation.spectral_variance: spectral variance.
Examples
from spafe.features.spfeats import extract_feats from scipy.io.wavfile import read from pprint import pprint # read audio fpath = "../../../tests/data/test.wav" fs, sig = read(fpath) # compute erb spectrogram spectral_feats = extract_feats(sig, fs) pprint(spectral_feats)
References