Interspeech 2016 - Experimental results for DNN-based clustering paper

The files in this dataset correspond to the results in the paper "DNN-based speaker clustering for speaker diarisation" published in Interspeech 2016. This paper proposes a new method of performing speaker diarisation using deep neural networks (DNNs) which involves replacing the final layer of the network and retraining in iterations. A stopping criterion is also proposed based on using posteriors as confidences.

The files are in the format '.rttm' which is the standard format for scoring a diarisation system. There is also a summary 'results' file which displays the numbers found in the paper.