2

I want to extract audio features using opensmile using the predefined emobase.config file. However when I enter a .wav and I get the output for only one framesize.

I changed the framemode to fixed and gave the values for fixedstep and fixedsize but it throws error. I don't know if the inbuilt config file like emobase.conf will generates output for the whole video and not frame wise.

Is there a way to extract features per frame size for inbuilt config files? I get frame wise output for self generated config files! However for the inbuilt ones I don't know how to go about it!

Soni Sol
  • 2,367
  • 3
  • 12
  • 23
Adrika
  • 51
  • 2

1 Answers1

2

This is caused by a setting in the configuration file. In the "data output configuration" at the end of the file, you will find "reader.dmLevel=fun". This means only functionals (measures taken over the whole sound file and not per frame) will be printed out. If you change it to "lld" (low level descriptors), it will print them out frame by frame. I think this applies to all emobase config files and the emo_large file as well.

Tl;dr -> change reader.dmLevel=fun to reader.dmLevel=lld in the data output configuration of your config file.

dthomas
  • 53
  • 4
  • 1
    Some comments to the above: The default setting is func, not fun. Setting it to lld will produce lots of numbers such as "0.000000e+00". Setting reader.dmLevel to lld;lld_de seems to produce valid results. – reggie Oct 18 '22 at 13:59