Multiple feature combination to improve speaker diarization of telephone conversations

Abstract

We report results on speaker diarization of telephone conversations. This speaker diarization process is similar to the multistage segmentation and clustering system used in broadcast news. It consists of an initial acoustic change point detection algorithm, iterative Viterbi re-segmentation, gender labeling, agglomerative clustering using a Bayesian information criterion (BIC), followed by agglomerative clustering using stateof-the-art speaker identification methods (SID) and Viterbi resegmentation using Gaussian mixture models (GMMs). The Viterbi re-segmentation using GMMs is new, and it reduces the diarization error rate (DER) by 10%. We repeat these multistage segmentation and clustering steps twice: once with MFCCs as feature parameters for the GMMs used in gender labeling, SID and Viterbi re-segmentation steps, and another time with Gaussianized MFCCs as feature parameters for the GMMs used in these three steps. The resulting clusters from the parallel runs are combined in a novel way that leads to a significant reduction in the DER. On a development set containing 30 telephone conversations, this combination step reduced the DER by 20%. On another test set containing 30 telephone conversations, this step reduced the DER by 13%. The best error rate we have achieved is 6.7% on the development set, and 9.0% on the test set.

Speaker diarization of French broadcast news

“…For telephone conversations, we showed in [8] that Viterbi resegmentation following SID clustering reduces the DER by 10%. We experimented with a similar module for broadcast news diarization.…”

A fast speaker indexing using vector quantization and second order statistics with adaptive threshold computation

“…For example, in [5] 12 dimensional Mel cepstral Frequency coefficient (MFCC) feature vectors are used and the value of the threshold λ is 12.0. In [12] 26 MFCCs (12 MFCCs + energy + their first derivatives) are used and λ was defined as 3.0. The computation of an adaptive threshold λ for unsupervised speaker segmentation using the BIC is presented in [13].…”