MeMo: Attentional Momentum for Real-time Audio-visual Speaker Extraction under Impaired Visual Conditions

TL;DR

This paper introduces MeMo, a real-time audio-visual speaker extraction framework that maintains attention momentum using adaptive memory banks, improving performance especially when visual cues are missing or degraded.

Contribution

The paper presents a novel attentional momentum framework with adaptive memory banks for robust real-time speaker extraction under impaired visual conditions.

Findings

Achieves at least 2 dB SI-SNR improvement over baseline.

Effective in scenarios with missing or degraded visual cues.

Maintains attention over time despite visual impairments.

Abstract

Audio-visual Target Speaker Extraction (AV-TSE) aims to isolate a target speaker's voice from multi-speaker environments by leveraging visual cues as guidance. However, the performance of AV-TSE systems heavily relies on the quality of these visual cues. In extreme scenarios where visual cues are missing or severely degraded, the system may fail to accurately extract the target speaker. In contrast, humans can maintain attention on a target speaker even in the absence of explicit auxiliary information. Motivated by such human cognitive ability, we propose a novel framework called MeMo, which incorporates two adaptive memory banks to store attention-related information. MeMo is specifically designed for real-time scenarios: once initial attention is established, the system maintains attentional momentum over time, even when visual cues become unavailable. We conduct comprehensive experiments to verify the effectiveness of MeMo. Experimental results demonstrate that our proposed framework achieves SI-SNR improvements of at least 2 dB over the corresponding baseline.