Large-Scale Speaker Diarization for Long Recordings and Small Collections

Abstract

Performing speaker diarization of very long recordings is a problem for most diarization systems that are based on agglomerative clustering with an HMM topology. Performing collectionwide speaker diarization, where each speaker is identified uniquely across the entire collection, is even a more challenging task. In this paper we propose a method with which it is possible to efficiently perform diarization of long recordings. We have also applied this method successfully to a collection of a total duration of approximately 15 hours. The method consists of first segmenting long recordings into smaller chunks on which diarization is performed. Next, a speaker detection system is used to link the speech clusters from each chunk and to assign a unique label to each speaker in the long recording or in the small collection. We show for three different audio collections that it is possible to perform high quality diarization with this approach. The long meetings from the ICSI corpus are processed 5.5 times faster than the originally needed time and by uniquely labeling each speaker across the entire collection it becomes possible to perform speaker-based information retrieval with high accuracy (mean average precision of 0.57).

Large-scale community detection on speaker content graphs

“…Each node is initialized as its own cluster and iteratively merged with other clusters via some similarity metric -we found the unweighted group average affinity to perform best [11] -until some stopping criterion (e.g., BIC [7,10], maximum distance [8], number of clusters, etc.) is met.…”

“…The related problem of largescale speaker diarization was also explored for long recordings and small collections in [8]. Both works solely consider the use of agglomerative hierarchical clustering; neither consider the community detection algorithms or the sparse graph structure we investigate in this paper.…”

Toeplitz Inverse Covariance Based Robust Speaker Clustering for Naturalistic Audio Streams

“…in a multispeaker audio stream [1]. Some of the practical applications of diarization technology include information retrieval [2], broadcast news, meeting conversations, telephone calls, VoIP, digital audio logging [3] and interaction analysis in Peer-Led Team Learning (PLTL) groups [4,5,6,7]. Diarization is a challenging task for naturalistic audio streams as they contain short conversational turns, overlapped speech, noise and reverberation [8,9].…”

Speaker diarization and linking of large corpora

“…Cluster and speaker purity measures are given to compare MAP and JFA approaches to adaptation. Targetting large scale speaker diarization is [5], which proposes a multi-stage system involving speaker diarization followed by speaker linking of chunks of speech data. Although splitting the database in small chunks increases diarization error rates, this system scales particurlarly well on large data sets.…”

“…Although splitting the database in small chunks increases diarization error rates, this system scales particurlarly well on large data sets. Only the work in [5] focuses on interview and meeting data, the others targetting telephone speech conversations between two people. In our work, we target a challenging scenario with meetings of 4 participants each recorded using various types of far-field microphones and several recording rooms.…”