Mutual information based channel selection for speaker diarization of meetings data

Abstract

In the meeting case scenario, audio is often recorded using Multiple Distance Microphones (MDM) in a non-intrusive manner. Typically a beamforming is performed in order to obtain a single enhanced signal out of the multiple channels. This paper investigates the use of mutual information for selecting the channel subset that produces the lowest error in a diarization system. Conventional systems perform channel selection on the basis of signal properties such as SNR, cross correlation. In this paper, we propose the use of a mutual information measure that is directly related to the objective function of the diarization system. The proposed algorithms are evaluated on the NIST RT 06 eval dataset. Channel selection improves the speaker error by 1.1% absolute (6.5% relative) w.r.t. the use of all channels.

Speech and Audio Processing for Coding, Enhancement and Recognition

“…Instead, for the aIB framework subsequent statistics are taken to be averages of previously defined statistics. Speaker diarization systems which employed aIB are predominantly implemented by Vijayasenan et al [140][141][142][143][144][145].…”

Selection of TDOA parameters for MDM speaker diarization

“…In [7] it has been presented a method to select the microphone pairs used to calculate TDOA's and the use of these TDOA's as the first stage of the segmentation and clustering module. Other alternative methods to select microphone pairs are presented in [8], [9], [10] and [11].…”

Analysis and development of robust speaker diarization for meetings

“…Still, the signal formed in this part will be used for the extraction of all the acoustic features and a very narrow selection could degrade the performance severely, without significant gain in computational time. As examples of this fourth group we could include the works in (Kumatani, McDonough, Lehman, & Raj, 2011) and (Vijayasenan, Valente, & Bourlard, 2009).  Another work, whose objective is to improve the composite signal, is presented in (Vijayasenan, Valente, & Bourlard, 2009).…”

“…As examples of this fourth group we could include the works in (Kumatani, McDonough, Lehman, & Raj, 2011) and (Vijayasenan, Valente, & Bourlard, 2009).  Another work, whose objective is to improve the composite signal, is presented in (Vijayasenan, Valente, & Bourlard, 2009). In this case the mutual information from N groups of channels is computed, and those with the highest mutual information are used in the beamforming stage.…”