0

I am a beginner of studying in Speaker verification.

I am studying about the DET curve!

Usually, papers mentioned that they used the normal deviate scale in the DET. But I failed to understand well about the normal deviate scale.

and the wikipedia only shows the normal deviate. I would like to know about the specific "normal deviate scale"!

and How can I transform from the x-axis of the ROC curve to the x-axis of the DET curve?

I really appreiciate in advance for your comment!

user2874612
  • 39
  • 1
  • 4

1 Answers1

0

This paper describes things pretty well:

http://www.itl.nist.gov/iad/mig/publications/storage_paper/det.pdf

You do not need to transform values, they remain the same. You just scale the axis in the graph when you plot it. Basically you need to shrink axis around 0 according to the special scale. In ROC curve distance between 1% and 2% is the same as the distance between 10% and 11%, in DET scale 1% and 2% take more on the graph.

Nikolay Shmyrev
  • 24,897
  • 5
  • 43
  • 87