Cantonese speech separation based on cross-modal feature fusion

Yuming Huang; Junfeng Wei; Zhenyuan Huang

doi:10.1117/12.3032932

5 July 2024 Cantonese speech separation based on cross-modal feature fusion

Yuming Huang, Junfeng Wei, Zhenyuan Huang

Proceedings Volume 13184, Third International Conference on Electronic Information Engineering and Data Processing (EIEDP 2024); 131841W (2024) https://doi.org/10.1117/12.3032932
Event: 3rd International Conference on Electronic Information Engineering and Data Processing (EIEDP 2024), 2024, Kuala Lumpur, Malaysia

Abstract

In most audio-visual speech separation models, audio and video features are generally directly merged. This approach may result in insufficient utilization of the interrelationships among cross-modal features. Therefore, joint attention is employed during the fusion of audio-visual features to achieve cross-modal integration. By jointly modeling both intra-modal and inter-modal relationships, features of each modality can focus on both another modality and themselves, thereby obtaining improved cross-modal fusion features to enhance speech separation performance.To evaluate the effectiveness of speech separation, an audio-visual bilingual Cantonese corpus (AVCC) was established, and the actual performance of three speech separation methods—Conv-TasNet, VisualVoice, and VisualVoice with joint cross-attention—was tested using Perceptual Evaluation of Speech Quality (PESQ), Short-Time Objective Intelligibility (STOI), and Signal-to-Distortion Ratio (SDR).The study indicates that when separating mixed speech of two speakers with different genders, the recommended VisualVoice method with joint cross-attention achieves an increase in SDR of at least 0.87dB, with the highest improvement reaching up to 1.99dB.The Joint Cross-Attention Model for feature fusion can utilize the correlations between cross-modalities more effectively, thereby enhancing the separation results.

(2024) Published by SPIE. Downloading of the abstract is permitted for personal use only.

Citation Download Citation

Yuming Huang, Junfeng Wei, and Zhenyuan Huang "Cantonese speech separation based on cross-modal feature fusion", Proc. SPIE 13184, Third International Conference on Electronic Information Engineering and Data Processing (EIEDP 2024), 131841W (5 July 2024); https://doi.org/10.1117/12.3032932

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE

Includes PDF, HTML & Video, when available

Members: $17.00

Non-members: $21.00 ADD TO CART

PROCEEDINGS
7 PAGES

DOWNLOAD PAPER SAVE TO MY LIBRARY

GET CITATION

RIGHTS & PERMISSIONS

Get copyright permission Get copyright permission on Copyright Marketplace

KEYWORDS

Visualization

Visual process modeling

Feature fusion

Performance modeling

Network architectures

Video

Motion analysis

Keywords/Phrases

Search In:

Publication Years