Speaker Diarization Features: The UPM Contribution to the RT09 Evaluation

Abstract

Abstract-Two new features have been proposed and used in the Rich Transcription Evaluation 2009 by the Universidad Politécnica de Madrid, which outperform the results of the baseline system. One of the features is the intensity channel contribution, a feature related to the location of the speaker. The second feature is the logarithm of the interpolated fundamental frequency. It is the first time that both features are applied to the clustering stage of múl-tiple distant microphone meetings diarization. It is shown that the inclusión of both features improves the baseline results by 15.36% and 16.71% relative to the development set and the RT 09 set, respectively. If we consider speaker errors only, the relative improvement is 23% and 32.83% on the development set and the RT09 set, respectively.

Selection of TDOA parameters for MDM speaker diarization

“…Most MDM systems use acoustic features as Mel-Frequency Cepstral Coefficients (MFCC) and localization features as the Time Delay Of Arrival (TDOA) values [1]. Other features used in some systems are the normalized energy of the channels [2] or the prosodic parameters [3] [4].…”

Towards an unsupervised speaking style voice building framework: multi.style speaker diarization

“…The system mush be able to robustly cope with noisy ASR-processed corpora and with challenging data such as interviews, debates, home recordings, political speeches, etc. The use of diarization techniques for speaker-turn segmentation will allow the system creating homogeneous voices from heterogeneous recordings, because the number of speakers would be automatically estimated in a fully unsupervised way, and language-independent diarization techniques automatically could provide the temporal labels of the turns of a certain speaker [2,3].…”

Speaker diarization in meetings domain

“…In the speech community, the Qualcomm-ICSI-OGI front end [21] is commonly used to perform Wiener filtering. Most state-of-the-art speaker diarization systems [10,11,15,22] applied Wiener filtering to all audio channels for speech enhancement before filtered and summed to produce a beamformed audio channel. In Van Leeuwen and Konecny [23], the filtering, however, was applied after beamforming.…”

“…The IAHC was initially proposed by Ajmera and Wooters [38] for speaker clustering and since then, the approach has been adopted by many others [11,15,22,23,178]. For completeness, this section provides the detail description of the IAHC clustering procedures.…”

“…The speech frames obtained from the output of the SAD module are uniformly grouped into K 0 clusters, where the value of K 0 is empirically determined from development data set, typically K 0 is from 15-20. Despite being such a naive strategy, many systems[12,15,22,66,78,150] have adopted this method and reported competitive results. Another variation of uniform initialization was suggested in Sun et al[10], where the segments from the SAD output were used.…”