Closed Menrath closed 11 months ago
The issue here is the preemphasis of the signal (higher frequencies are amplified) employed in the emo_large.conf
config file, so that the spectrum is not "white" anymore.
If you change line 81 https://github.com/audeering/opensmile/blob/341aea9ce52bc63d7fe75098027a394994f8493b/config/misc/emo_large.conf#L81C1-L81C24 to
reader.dmLevel=frames
you will get the expected result.
Generally, you might want to try also ComParE_2016.conf
, which does not use preemphasis and has been updated more recently:
https://github.com/audeering/opensmile/blob/master/config/compare16/ComParE_2016.conf
When sampling at 48kHz the spectral centroid of a white noise signal should be in theory (also quite intuitively) half of the Nyquist frequency (24kHz in that case) leading to 12kHz.
OpenSMILE calculated a value about 168xxHz. I tested this using the packaged emo_large.conf.
To reproduce one can use this python script:
which prints: