Topics

change in c0 values from To MFCC... after Praat upgrade

Evanini, Keelan
 

Hi all,

I recently upgraded from Praat version 5.2.38 to version 6.0.38 and noticed that the c0 (energy) values produced by the To MFCC... command are very different in the new version. For example, I ran the following script on the same audio file using the two different versions of Praat.

--------------------------------------------------------------
form Get_arguments
text wavfile
text outfile
endform

Read from file... 'wavfile$'

To MFCC... 12 0.025 0.01 100.0 100.0 0.0

Write to text file... 'outfile$'
--------------------------------------------------------------

The following excerpts from the output files demonstrate that the c0 values are quite different while the values of the 12 coefficients are nearly identical.

1) output using Praat v. 5.2.38

File type = "ooTextFile"
Object class = "MFCC"

xmin = 0
xmax = 9.350875
nx = 931
dx = 0.01
x1 = 0.025437499999999887
fmin = 100
fmax = 2700
maximumNumberOfCoefficients = 12
frame []:
frame [1]:
numberOfCoefficients = 12
c0 = 49.11119872695355
c []:
c [1] = 15.468436129271204
c [2] = 45.916023265145945
c [3] = 45.79296277254156
c [4] = 19.272906660022414
c [5] = 32.86717561439363
c [6] = 16.845067539875437
c [7] = 7.262801697852275
c [8] = 3.4172383528865495
c [9] = 29.99230975898888
c [10] = 29.303995186370344
c [11] = -2.432205289932938
c [12] = -1.877080921494958

...

frame [931]:
numberOfCoefficients = 12
c0 = 50.2330942215624
c []:
c [1] = 76.46967854810613
c [2] = 49.19088677884566
c [3] = 77.85698084678515
c [4] = 39.29269492580826
c [5] = 55.24431135389732
c [6] = 36.187809377893515
c [7] = 36.41939156999201
c [8] = 26.350869954840622
c [9] = 37.31959362118782
c [10] = 20.468665684425353
c [11] = 20.331237619467593
c [12] = 23.172112339794772

2) output using Praat v. 6.0.38

File type = "ooTextFile"
Object class = "MFCC 1"

xmin = 0
xmax = 9.350875
nx = 931
dx = 0.01
x1 = 0.025437499999999887
fmin = 0
fmax = 2800
maximumNumberOfCoefficients = 26
frame []:
frame [1]:
numberOfCoefficients = 12
c0 = 1326.0023656277458
c []:
c [1] = 15.468436129271254
c [2] = 45.91602326514581
c [3] = 45.79296277254148
c [4] = 19.272906660022343
c [5] = 32.867175614393396
c [6] = 16.845067539875522
c [7] = 7.26280169785209
c [8] = 3.417238352886393
c [9] = 29.992309758988746
c [10] = 29.303995186370955
c [11] = -2.4322052899326536
c [12] = -1.8770809214949224

...

frame [931]:
numberOfCoefficients = 12
c0 = 1356.293543982185
c []:
c [1] = 76.46967854810603
c [2] = 49.190886778845574
c [3] = 77.85698084678504
c [4] = 39.29269492580821
c [5] = 55.24431135389703
c [6] = 36.18780937789363
c [7] = 36.41939156999184
c [8] = 26.350869954840466
c [9] = 37.319593621187686
c [10] = 20.468665684425968
c [11] = 20.33123761946785
c [12] = 23.17211233979486

Has the procedure for calculating the c0 value changed in more recent versions of Praat?

In addition, I notice that regardless of the values that I specify for the "Number of coefficients" and "Position of first filter" arguments to the To MFCC... command, the header in the output always contains the following values using Praat v. 6.0.38 (whereas the values in the header of the output file vary as expected using Praat 5.2.38 based on the values of the arguments specified by the users):

fmin = 0
fmax = 2800
maximumNumberOfCoefficients = 26

I don't know whether this is related to the observed differences in the c0 values or not, and it doesn't appear that the values in the header are the ones actually being used to compute the MFCCs (since the MFCC values correspond to the ones computed by the earlier version of Praat and the appropriate number of coefficients are listed in the output file for each frame), but I thought I should point it out, since it is unexpected behavior.

I have observed this behavior using Praat for Linux, Windows, and Mac. Does anyone have any advice about why this might be happening?

Thanks,
Keelan


________________________________

This e-mail and any files transmitted with it may contain privileged or confidential information. It is solely for use by the individual for whom it is intended, even if addressed incorrectly. If you received this e-mail in error, please notify the sender; do not disclose, copy, distribute, or take any action in reliance on the contents of this information; and delete it from your system. Any other use of this e-mail is prohibited.


Thank you for your compliance.

________________________________