Speaker Embedding Informed Audiovisual Active Speaker Detection for Egocentric Recordings

TL;DR

This paper introduces SCAN, a speaker comparison network that enhances audiovisual active speaker detection in egocentric recordings by leveraging speaker embeddings from reference speech, improving accuracy in noisy, dynamic scenes.

Contribution

The paper presents SCAN, a novel auxiliary network utilizing speaker-specific information to disambiguate challenging scenes in active speaker detection, especially in egocentric recordings.

Findings

SCAN improves mAP by 14.5% with TalkNet

SCAN improves mAP by 10.3% with Light-ASD

Enhanced speaker embedding utilization boosts detection accuracy

Abstract

Audiovisual active speaker detection (ASD) addresses the task of determining the speech activity of a candidate speaker given acoustic and visual data. Typically, systems model the temporal correspondence of audiovisual cues, such as the synchronisation between speech and lip movement. Recent work has explored extending this paradigm by additionally leveraging speaker embeddings extracted from candidate speaker reference speech. This paper proposes the speaker comparison auxiliary network (SCAN) which uses speaker-specific information from both reference speech and the candidate audio signal to disambiguate challenging scenes when the visual signal is unresolvable. Furthermore, an improved method for enrolling face-speaker libraries is developed, which implements a self-supervised approach to video-based face recognition. Fitting with the recent proliferation of wearable devices, this work focuses on improving speaker-embedding-informed ASD in the context of egocentric recordings, which can be characterised by acoustic noise and highly dynamic scenes. SCAN is implemented with two well-established baselines, namely TalkNet and Light-ASD; yielding a relative improvement in mAP of 14.5% and 10.3% on the Ego4D benchmark, respectively.