Improvements to PLDA i-vector scoring for short segments clustering

This paper extends upon previous work using Mean Shift algorithm to perform speaker clustering on i-vectors generated from short speech segments. In this paper we examine the effectiveness of Spherical Normalization in the presence of different numbers of speakers. This normalization method is not only easy to implement but also improves clustering results. The main improvement is that the proposed mean shift algorithm is more robust to changes in the number of speakers. In the case of 30 speakers, we achieved evaluation parameter K of 75.3 compared to 72.1 with the baseline system.