How do I tune the bob engine so as to lower the errors?

[Maria Garcia]

Hi,

I first want to thank everyone to helping me with the issues I had before in trying to get the Bob engine working.  I still don't have all parts working but it is moving forward now and I can get through a full run. 

I ran a couple thousand tests and came up with the following curve (it is attached).  However the miss/false alarm rates were high.  what can I do to bring the rates closer to the zero mark?  I'm using ivector and I was using the cur_type flag ==> lnorm_ivector  (using spkverif_ivector.py)

I'm using voxforge.py calling mechanism  but overriding with my own data because when I tried using the standard file verification approach I kept getting a seg fault.

As far as I can tell voxforge.py also calls the standard file verification approach but I'm guessing does something a little bit differently to prevent the crash.

I'm using bob 1.17  .  I'm using out of the box configuration settings  (except for where my data is stored).

Thank you for any help.