maad.sound.spectrum

maad.sound.spectrum(s, fs, nperseg=256, noverlap=None, nfft=None, window='hann', method='welch', tlims=None, flims=None, scaling='spectrum', as_pandas_series=False, display=False)[source]

Estimate the power spectral density or power spectrum of 1D signal.

The estimates can be computed using two methods: Welch or periodogram. Welch’s method divides the signal into segments, computes the power spectral density for each segment and then takes the average between segments. The periodogram method computes the power spectral density of the input signal using a defined window (Hanning window by default).

Parameters:

s: 1D array: Input signal to process
fs: float, optional: Sampling frequency of audio signal
nperseg: int, optional: Length of segment for ‘welch’ method, default is 256
noverlap: int, optional: Overlap between segments for Welch’s method. If None, noverlap = nperseg/2.
windowstring, default is ‘hann’: Name of the window used for the short fourier transform, ‘hanning’ window is now deprecated.
nfft: int, optional: Length of FFT for periodogram method. If None, length of signal will be used. Length of FFT for welch method if zero padding is desired. If None, length of nperseg will be used.
method: {‘welch’, ‘periodogram’}: Method used to estimate the power spectral density of the signal
tlims: tuple of ints or floats: Temporal limits to compute the power spectral density in seconds (s) If None, estimates for the complete signal will be computed. Default is ‘None’
flims: tuple of ints or floats: Spectral limits to compute the power spectral density in Hertz (Hz) If None, estimates from 0 to fs/2 will be computed. Default is ‘None’
scaling: {‘spectrum’, ‘density’}, optional: Choose between power spectrum (units V**2) or power spectral density (units V**2/Hz) scaling. Defaults to ‘spectrum’.
as_pandas_series: bool: Return data as a pandas.Series. This is usefull when computing multiple features over a signal. Default is False.

Returns:

pxx: pandas Series: Power spectral density estimate.
f_idx: pandas Series: Index of sample frequencies.

Notes

This is a wrapper that uses functions from Scipy (scipy.org), in particular from the scipy.signal module.

Examples

>>> from maad import sound
>>> s, fs = sound.load('../data/spinetail.wav')
>>> spec, f_idx = sound.spectrum(s, fs, nperseg=512, display=True)

Specify temporal and spectral limits to get spectral characteristics of the spinetail’s song.

>>> spec, f_idx = sound.spectrum(s, fs, nperseg=2056, tlims=(5.3, 7.9), flims=(2000, 12000), display=True)