Combine MLLR adaptation with MAP adaptation

[Toan Nguyen]

Dear cmusphinx,

Could you please tell us how to combine MLLR adaptation with MAP adaptation
that was already mentioned from the link below

http://cmusphinx.sourceforge.net/wiki/tutorialadapt

Yours sincerely, Toan

[Nickolay V. Shmyrev]

1. Run bw to collect statistics
2. Estimate mllr transform
3. Update means with mllr
4. Run bw again with updated means
5. Apply MAP adaptation with fixed tau greater than 100 (try to select the best value). Unfortunately from my experience automatic tau selection is broken in map_adapt. This way you'll update the variances a bit, but only slightly.

[Nickolay V. Shmyrev]

http://nshmyrev.blogspot.de/2009/09/adaptation-methods.html

[Balaji]

Hello,

I am refering to the 3rd step "Update means with mllr". Why are variances not updated?
"Mean and Variance Adaptation within the MLLR Framework", by Gales and Woodland says, "To more closely model the data in speaker adaptation, the gaussian variances also should be modified". Can you please clarify.
Thank you.

[Nickolay V. Shmyrev]

Variance update is simply not yet implemented. It is implemented in more advanced toolkits.

[Balaji]

Thank you.

[Toan Nguyen]

Okay thanks, I will give it a try!

[Toan Nguyen]

Dear Mr. Nickolay,

I guess the step #3 "Update means with mllr" is "Apply it to the baseline means with
mllr_transform" the one was already mentioned in the document below

http://www.cs.cmu.edu/~archan/presentation/MAP.pdf

Please tell us how how we use the tool mllr_transform and any paramters we should pass to.

Yours sincerely, Toan

[Toan Nguyen]

I found it answered from the link below

http://www.cs.cmu.edu/~archan/documentation/mllr_transform.pdf

Help and example:
nmllr_transform -help yes -example yes

[Balaji]

Hello,

I have used the details in this discussion in my project. In the 5th step (Apply MAP adaptation with fixed tau greater than 100 (try to select the best value).....):

How to set tau value?

Thank you.