Version 0.2.0

• The code has been extensively restructured and is now much more modular, with most of the closure, release, and voicing onset detection methods stored in separate functions.
• A new voicing onset detection method has been added using strength of harmonic excitation (vo_method='soe'). This is now the default for predicting negative VOT, as it works quite well in the tests I've run and is much faster than all other methods. For this reason, the when_avoid_soe argument has been added to neg_setParams(), specifying an error value where other methods should be tried (default is 10 ms), since neg_setParams() can otherwise be very slow.
• A new burst detection method has been added, where the speech signal is differenced before searching for sudden amplitude increases in the differenced signal. This is now the default when looking for positive VOT, as it works quite well in the tests I've run.
• In order to speed up negativeVOT I've removed part of the functionality for estimating whether a stop has prevoicing or not (the existing functionality didn't work very well, and it wasn't really worth the speed costs). You can still specify that getVOT() should look for both positive and negative VOT, but it'll likely perform even worse than before. I haven't found a solution to this yet.
• The line widths in plots produced by getVOT has been increased, and a legend is now included when using plot_training_diff().

Version 0.1.4

Various bugs fixed since the last release. Most important update is that addVOT2emuDB() is now done completely differently -- it doesn't rely on getting the track data with emuR but instead loads in the wav files directly, which is much quicker and more flexible it turns out.

Version 0.1.1

Added zero crossing rate as a method for voicing detection