API reference for LPC?

[akhil2495]

Can parselmouth be used to extract linear predictive coding coefficients?

[YannickJadoul]

Yes, but there's indeed no API reference for this, because there is no Python API, yet.

Until I add this (or someone else does; help welcome ;-) ), you can go through parselmouth.praat.call:

>>> import parselmouth
>>> s = parselmouth.Sound("the_north_wind_and_the_sun.wav")
>>> lpc = parselmouth.praat.call(s, "To LPC (autocorrelation)", 16, 0.025, 0.005, 50.0)
>>> lpc
<parselmouth.Data object at 0x7f4f3f5d3c00>
>>> lpc.class_name
'LPC'
>>> print(lpc)
Object type: LPC
Object name: untitled
Date: Thu Nov 22 11:38:48 2018

Time domain: 0 to 1.283265306122449 (s).
Prediction order: 16
Number of frames: 247
Time step: 0.005 (s).
First frame at: 0.026632653061224473 (s).

As you can see, the Python type of lpc is parselmouth.Data (so it does not have a lot of methods), but the actual underlying Praat type is "LPC", which means you can use it in further calls to parselmouth.praat.call until you get a more useful object or the values you're interested in. For example, you could:

>>> parselmouth.praat.call(lpc, "Get number of coefficients", 1)
16

Or to get all raw values (consult the Praat documentation to see what these actually mean!), you could:

>>> lpc_matrix = parselmouth.praat.call(lpc, "Down to Matrix (lpc)")
>>> lpc_matrix
<parselmouth.Matrix object at 0x7f4f3f5eac00>
>>> lpc_matrix.values
array([[ 0.12599566,  0.12089129,  0.1061873 , ..., -1.19781365,
        -1.14358651, -1.1659433 ],
       [ 0.51795006,  0.46722564,  0.51150008, ...,  1.14546318,
         1.09264874,  1.14384542],
       [-0.01884895,  0.02674646,  0.04591352, ..., -1.38964444,
        -1.42701829, -1.50292197],
       ...,
       [-0.01957717, -0.00318544, -0.09061003, ...,  0.46313035,
         0.41949584,  0.54231048],
       [-0.06745527, -0.0395186 , -0.03539812, ..., -0.2470586 ,
        -0.23618874, -0.33601743],
       [-0.01429427,  0.00990911, -0.02896412, ...,  0.05589769,
         0.06509704,  0.10742232]])
>>> lpc_matrix.values.shape
(16, 247)

I'll have a look how much work it would be to port LPC to the Python API, and see if I can maybe find some time within the next few months, so you can leave this issue open for now.