Caution

You're reading the documentation for a development version. For the latest released version, please have a look at 0.11.0.

librosa.effects.percussive

librosa.effects.percussive(y, *, kernel_size=31, power=2.0, mask=False, margin=1.0, n_fft=2048, hop_length=None, win_length=None, window='hann', center=True, pad_mode='constant')[source]

Extract percussive elements from an audio time-series.

Parameters:
ynp.ndarray [shape=(…, n)]

audio time series. Multi-channel is supported.

kernel_sizeint or tuple (kernel_harmonic, kernel_percussive)

Kernel size(s) for the median filters.

  • If scalar, the same size is used for both harmonic and percussive.

  • If tuple, the first value specifies the width of the harmonic filter, and the second value specifies the width of the percussive filter.

powerfloat > 0 [scalar]

Exponent for the Wiener filter when constructing soft mask matrices.

maskbool

Return the masking matrices instead of components.

marginfloat or tuple (margin_harmonic, margin_percussive)

Margin size(s) for the masks.

  • If scalar, the same size is used for both harmonic and percussive.

  • If tuple, the first value specifies the margin of the harmonic mask, and the second value specifies the margin of the percussive mask.

n_fftint > 0 [scalar]

Length of the windowed signal after padding with zeros. The number of rows in the STFT matrix is (1 + n_fft/2).

hop_lengthint or None

Number of audio samples between adjacent STFT columns. If unspecified, defaults to win_length // 4.

win_lengthint or None

Each frame of audio is windowed by window of length win_length and then padded with zeros to match n_fft. If unspecified, defaults to win_length = n_fft.

windowstr, tuple, number, function, or np.ndarray [shape=(n_fft,)]

Window specification. See scipy.signal.get_window for supported values.

centerbool

If True, the signal is padded so that frame t is centered at y[t * hop_length].

pad_modestr

Padding mode used when center=True. See numpy.pad for available modes.

Returns:
y_percussivenp.ndarray [shape=(…, n)]

audio time series of just the percussive portion

See also

hpss

Separate harmonic and percussive components

harmonic

Extract only the harmonic component

librosa.decompose.hpss

HPSS for spectrograms

Examples

>>> # Extract percussive component
>>> y, sr = librosa.loadx('choice')
>>> y_percussive = librosa.effects.percussive(y)

>>> # Use a margin > 1.0 for greater percussive separation
>>> y_percussive = librosa.effects.percussive(y, margin=3.0)