SPEAR - disabling ZT-norm

[Neve Stagy]

Hello,

I'm running some experiments using SPEAR's GMM-UBM for speaker verification. I want to rely entirely on log likelihood for scoring (as mentioned in Reynolds et. al on which this module seems based off of), rather than ZT-norm scoring.

I noticed older versions of SPEAR included a "--no-zt-norm" option. Does the current version include anything similar? If not, how may I disable it?

Best,

-Stefan

[Neve Stagy]

Never mind, I noticed it "--zt-norm' is False by default. 

Can anyone please elaborate on how GMM-UBM scores are computed? I'm running some optimization algorithms which achieve ideal LLR scores, but are receiving strangely low scores in SPEAR's results.

[Amir Mohammadi]

Does this:
https://www.idiap.ch/software/bob/docs/bob/bob.learn.em/stable/guide.html#linear-scoring

help?

Best,

Amir

--
-- You received this message because you are subscribed to the Google Groups bob-devel group. To post to this group, send email to bob-...@googlegroups.com. To unsubscribe from this group, send email to bob-devel+...@googlegroups.com. For more options, visit this group at https://groups.google.com/d/forum/bob-devel or directly the project website at http://idiap.github.com/bob/
---
You received this message because you are subscribed to the Google Groups "bob-devel" group.
To unsubscribe from this group and stop receiving emails from it, send an email to bob-devel+...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.