Dynamic Bayesian networks for meeting structuring

Abstract

This paper is about the automatic structuring of multiparty meetings using audio information. We have used a corpus of 53 meetings, recorded using a microphone array and lapel microphones for each participant. The task was to segment meetings into a sequence of meeting actions, or phases. We have adopted a statistical approach using dynamic Bayesian networks (DBNs). Two DBN architectures were investigated: a two-level hidden Markov model (HMM) in which the acoustic observations were concatenated; and a multistream DBN in which two separate observation sequences were modelled. Additionally we have also explored the use of counter variables to constrain the number of action transitions. Experimental results indicate that the DBN architectures are an improvement over a simple baseline HMM, with the multistream DBN with counter constraints producing an action error rate of 6%.

Automatic Meeting Segmentation Using Dynamic Bayesian Networks

“…Previously, we have outlined a meeting action recognition framework based on acoustic and lexical related features and a layered multistream dynamic Bayesian network model [19], [21]. This model combines the advantages of independent feature-stream processing together with a structured approach.…”

“…The vector consists of all possible products of the six sound activity locations during a time window of three frames [19] where each vector highlights the turn taking interaction pattern around the time . Considering, for simplicity, a smaller turn taking matrix evaluated only on two frames the diagonal elements highlight whether a speaker active at time , is still speaking at time .…”

Analyzing Group Interactions in Conversations: a Review

“…Dielmann et al [15] proposed two approaches for meeting structuring from audio-only features using multilevel Dynamic Bayesian Networks (DBNs). The first DBN decomposed the group activities as sequences of sub-actions with no explicit meaning.…”

Layered HMM for Motion Intention Recognition

“…Hierarchical Hidden Markov Models (HHMMs) and layered hidden Markov models have been used to model various phenomena that exhibit stochastic structures at several different levels in areas such as speech and text recognition, modeling of group actions in meetings and extracting context from video, [12], [15]- [18]. Zhang et al used a two-layer HMM to model individual and group actions during meetings in [16].…”