LSTMSE-Net: Long Short Term Speech Enhancement Network for Audio-visual Speech Enhancement

TL;DR

This work addresses the challenge of robust speech enhancement under adverse acoustics by integrating audio with lip-reading visual cues through an AVSE framework. The proposed LSTMSE-Net combines VisualFeatNet visual features with an audio encoder/decoder and uses a multi-modal, LSTM-based separator to produce a masking representation that suppresses noise before reconstruction. Empirical results on a LRS3-based AVSE dataset show LSTMSE-Net outperforms the COG-MHEAR 2024 baseline across PESQ, STOI, and SISDR, while using significantly fewer parameters (~5.1M vs ~75M) and achieving faster inference (~0.3 s per video). The architecture demonstrates strong efficiency and scalability, making real-time AVSE more feasible, with future directions including causality for real-time deployment and attention-based feature fusion.

Abstract

In this paper, we propose long short term memory speech enhancement network (LSTMSE-Net), an audio-visual speech enhancement (AVSE) method. This innovative method leverages the complementary nature of visual and audio information to boost the quality of speech signals. Visual features are extracted with VisualFeatNet (VFN), and audio features are processed through an encoder and decoder. The system scales and concatenates visual and audio features, then processes them through a separator network for optimized speech enhancement. The architecture highlights advancements in leveraging multi-modal data and interpolation techniques for robust AVSE challenge systems. The performance of LSTMSE-Net surpasses that of the baseline model from the COG-MHEAR AVSE Challenge 2024 by a margin of 0.06 in scale-invariant signal-to-distortion ratio (SISDR), $0.03$ in short-time objective intelligibility (STOI), and $1.32$ in perceptual evaluation of speech quality (PESQ). The source code of the proposed LSTMSE-Net is available at \url{https://github.com/mtanveer1/AVSEC-3-Challenge}.

Figures (1)

• Figure 1: (a) The workflow of the proposed LSTMSE-Net, (b) a single unit in the Separator block of the proposed LSTMSE-Net.