In speaker recognition, to train LDA and PLDA we need labeled developement data, if we don't have labeled developement data for training LDA and PLDA, how we train the LDA and PLDA, and which labeled database is perfect for this purpose?
Stanisław Kacprzak
You could try to cluster development data first and then use the clusters as labels.
Check this:
https://groups.csail.mit.edu/sls/publications/2014/shum-interspeech14.pdf
Also, this approach was explored in papers connected with NIST i-vector challenge 2014.
Check:
http://cs.uef.fi/odyssey2014/program/pdfs/25.pdf and other papers from Oddyssey 2014
0 votes
0 thanks
- Badges
- Science method
More Rizwan Ishaq's questions See All
Similar questions and discussions