librosa / librosa / 380

99%

Pull #235

travis-ci

Add squared window normalization to istft

Squared window normalization is mathmatically derived explicitly so that
minimizing the mean squared error between STFT of the estimated signal and STFT
of a modified or unmodified STFT as described in the Griffin's paper.
In the case of unmodified STFT, this leads perfect reconstrunction in reasonable
analysis and synthesis settings.

In this commit,

- remove unnecesarry magic window scaling in istft
- add tests that ensures istft is able to reconstruct signals with a fairly wide variety of n_fft, hop_length and window functions (in scipy.signal)
- remove trailing whitespaces (my editor automatically does)

The algorithm is based on the following paper:
D. W. Griffin and J. S. Lim, "Signal estimation from modified short-time Fourier transform,"
IEEE Trans. ASSP, vol.32, no.2, pp.236–243, Apr. 1984.
http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.306.7858

As for the tests:

1) white noise 2) sine wave, 3) real music signal are now covered. Also I removed
broken `test_istft`, since new `test_istft_reconstrucntion` ensures that istft is
able to reconstruct signals precisely with a fairly wide variety of n_fft, hop_length
and window functions, which were not tested enough previously.

3 of 3 new or added lines in 1 file covered. (100.0%)

1932 of 1945 relevant lines covered (99.33%)

1.99 hits per line