For installation and other information, see the Kaldi-trunk
Branches in this fork of Kaldi contain code for different papers, as listed below.
-
Probing the information encoded in x-vectors, Desh Raj, David Snyder, Daniel Povey, and Sanjeev Khudanpur. IEEE ASRU 2019: Check out the reddots branch and look under egs/reddots/v1.
-
Multi-class spectral clustering with overlaps for speaker diarization. Desh Raj, Zili Huang, and Sanjeev Khudanpur. IEEE SLT 2021. Check out the slt21_spectral branch and look under the files changed in this commit. The actual overlap-aware SC code can be found in my fork of scikit-learn. Detailed description of the code is available here.
If you use any of the above code, please cite the respective papers.
For any questions, please contact me at [email protected] or [email protected].