C-DPCL: an adaptive conformer-based source separation method on deep clustering

Shunxian Gu

doi:10.1117/12.2656634

20 October 2022 C-DPCL: an adaptive conformer-based source separation method on deep clustering

Shunxian Gu

Proceedings Volume 12451, 5th International Conference on Computer Information Science and Application Technology (CISAT 2022); 1245155 (2022) https://doi.org/10.1117/12.2656634
Event: 5th International Conference on Computer Information Science and Application Technology (CISAT 2022), 2022, Chongqing, China

Abstract

The cocktail-party problem is to separate each source from an overlapped speech. Transformer-based source separation models have achieved impressive performance in single-channel scenarios. This paper proposes using Conformer models on deep clustering approaches in clean and noisy environments. First, we replace the Bi-LSTM-based embedding network in the original deep clustering work with a Conformer structure. Second, canopy K-means is applied as the postprocessing module to adapt our model to cases with an unknown speakers’ number. In addition to tuning the model structure, we also introduce a denoising front-end to enable our model to handle noisy mixture signals. Experimental results on WSJ0-2mix show that our model possesses state-of-the-art performance among previous works in clean source separation. In contrast, on handling noisy source signals, experiments on LibriMix demonstrate the effectiveness of our ensemble model, with an average 6% increase in SDRi, compared with previous SOTA models.

Citation Download Citation

Shunxian Gu "C-DPCL: an adaptive conformer-based source separation method on deep clustering", Proc. SPIE 12451, 5th International Conference on Computer Information Science and Application Technology (CISAT 2022), 1245155 (20 October 2022); https://doi.org/10.1117/12.2656634

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
8 PAGES

DOWNLOAD PAPER SAVE TO MY LIBRARY

GET CITATION

RIGHTS & PERMISSIONS

Get copyright permission Get copyright permission on Copyright Marketplace

KEYWORDS

Data modeling

Neural networks

Binary data

Interference (communication)

Computer programming

Denoising

Time-frequency analysis

Show All Keywords

Keywords/Phrases

Search In:

Publication Years