import numpy
from numpy.lib.stride_tricks import sliding_window_view
[docs]
def stft(samples, framesize, hopsize):
"""
Estimate the DFT matrix for the given sample array.
Parameters
----------
samples : ndarray, list, float
Array of samples.
Returns
-------
dfts : ndarray
DFT matrix of shape (samples,frequencies).
"""
samples = numpy.atleast_1d(samples)
assert samples.ndim == 1, f'Expected 1D array (samples,), got {samples.shape}!'
frames = sliding_window_view(samples, framesize, writeable=False)[::hopsize]
dfts = numpy.zeros((len(frames), len(numpy.fft.rfftfreq(framesize))), complex)
w = 0.5 - 0.5 * numpy.cos(2 * numpy.pi * numpy.arange(framesize) / framesize)
for i, frame in enumerate(frames):
dfts[i] = numpy.fft.rfft(w * frame, norm='forward')
return dfts
[docs]
def istft(dfts, framesize, hopsize):
"""
Synthesize the sample array from the given DFT matrix.
Parameters
----------
dfts : ndarray
DFT matrix of shape (samples,frequencies).
Returns
-------
samples : ndarray
Array of samples.
"""
dfts = numpy.atleast_2d(dfts)
assert dfts.ndim == 2, f'Expected 2D array (samples,frequencies), got {dfts.shape}!'
samples = numpy.zeros((len(dfts) * hopsize + framesize), float)
frames = sliding_window_view(samples, framesize, writeable=True)[::hopsize]
w = 0.5 - 0.5 * numpy.cos(2 * numpy.pi * numpy.arange(framesize) / framesize)
w *= hopsize / numpy.sum(w**2) # unity gain
for i, dft in enumerate(dfts):
frames[i] += w * numpy.fft.irfft(dft, norm='forward')
return samples