Deep Neural Network Based Low-Latency Speech Separation with Asymmetric Analysis-Synthesis Window Pair

Shanshan Wang; Gaurav Naithani; Archontis Politis; Tuomas Virtanen

doi:10.23919/EUSIPCO54536.2021.9616165

Deep Neural Network Based Low-Latency Speech Separation with Asymmetric Analysis-Synthesis Window Pair

Shanshan Wang, Gaurav Naithani, Archontis Politis, Tuomas Virtanen

Computing Sciences

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Scientific › peer-review

2 Downloads (Pure)

Abstract

Time-frequency masking or spectrum prediction computed via short symmetric windows are commonly used in low-latency deep neural network (DNN) based source separation. In this paper, we propose the use of an asymmetric analysis-synthesis window pair which allows for training with targets with better frequency resolution, while retaining the low-latency during inference suitable for real-time speech enhancement or assisted hearing applications. In order to assess our approach across various model types and datasets, we evaluate it with a speaker-independent deep clustering (DC) model and a speaker-dependent mask inference (MI) model. We report an improvement in separation performance of up to 1.5 dB in terms of source-to-distortion ratio (SDR) while maintaining an algorithmic latency of 8 ms.

Original language	English
Title of host publication	European Signal Processing Conference 2021
Publisher	IEEE
Pages	301-305
Number of pages	5
ISBN (Electronic)	978-9-0827-9706-0
DOIs	https://doi.org/10.23919/EUSIPCO54536.2021.9616165
Publication status	Published - 2021
Publication type	A4 Article in conference proceedings
Event	European Signal Processing Conference - Dublin, Ireland Duration: 23 Aug 2021 → 27 Aug 2021 https://eusipco2021.org

Publication series

Name	European Signal Processing Conference
ISSN (Electronic)	2076-1465

Conference

Conference	European Signal Processing Conference
Abbreviated title	EUSIPCO
Country/Territory	Ireland
City	Dublin
Period	23/08/21 → 27/08/21
Internet address	https://eusipco2021.org

Keywords

Deep learning
Training
Time-frequency analysis
Source separation
Signal processing algorithms
Europe
Speech enhancement
Monaural speaker separation
Low latency
Asymmetric windows
Deep clustering

Publication forum classification

Publication forum level 1

Access to Document

10.23919/EUSIPCO54536.2021.9616165

2106.11794Accepted author manuscript, 3.62 MB

Cite this

@inproceedings{86bf8c71edb8469ca2e57a6b4d5cba81,

title = "Deep Neural Network Based Low-Latency Speech Separation with Asymmetric Analysis-Synthesis Window Pair",

abstract = "Time-frequency masking or spectrum prediction computed via short symmetric windows are commonly used in low-latency deep neural network (DNN) based source separation. In this paper, we propose the use of an asymmetric analysis-synthesis window pair which allows for training with targets with better frequency resolution, while retaining the low-latency during inference suitable for real-time speech enhancement or assisted hearing applications. In order to assess our approach across various model types and datasets, we evaluate it with a speaker-independent deep clustering (DC) model and a speaker-dependent mask inference (MI) model. We report an improvement in separation performance of up to 1.5 dB in terms of source-to-distortion ratio (SDR) while maintaining an algorithmic latency of 8 ms.",

keywords = "Deep learning, Training, Time-frequency analysis, Source separation, Signal processing algorithms, Europe, Speech enhancement, Monaural speaker separation, Low latency, Asymmetric windows, Deep clustering",

author = "Shanshan Wang and Gaurav Naithani and Archontis Politis and Tuomas Virtanen",

note = "jufoid=55867; European Signal Processing Conference, EUSIPCO ; Conference date: 23-08-2021 Through 27-08-2021",

year = "2021",

doi = "10.23919/EUSIPCO54536.2021.9616165",

language = "English",

series = "European Signal Processing Conference",

publisher = "IEEE",

pages = "301--305",

booktitle = "European Signal Processing Conference 2021",

url = "https://eusipco2021.org",

}

Wang, S , Naithani, G , Politis, A & Virtanen, T 2021, Deep Neural Network Based Low-Latency Speech Separation with Asymmetric Analysis-Synthesis Window Pair. in European Signal Processing Conference 2021. European Signal Processing Conference, IEEE, pp. 301-305, European Signal Processing Conference, Dublin, Ireland, 23/08/21. https://doi.org/10.23919/EUSIPCO54536.2021.9616165

Deep Neural Network Based Low-Latency Speech Separation with Asymmetric Analysis-Synthesis Window Pair. / Wang, Shanshan ; Naithani, Gaurav ; Politis, Archontis et al.
European Signal Processing Conference 2021. IEEE, 2021. p. 301-305 (European Signal Processing Conference).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Scientific › peer-review

TY - GEN

T1 - Deep Neural Network Based Low-Latency Speech Separation with Asymmetric Analysis-Synthesis Window Pair

AU - Wang, Shanshan

AU - Naithani, Gaurav

AU - Politis, Archontis

AU - Virtanen, Tuomas

N1 - jufoid=55867

PY - 2021

Y1 - 2021

N2 - Time-frequency masking or spectrum prediction computed via short symmetric windows are commonly used in low-latency deep neural network (DNN) based source separation. In this paper, we propose the use of an asymmetric analysis-synthesis window pair which allows for training with targets with better frequency resolution, while retaining the low-latency during inference suitable for real-time speech enhancement or assisted hearing applications. In order to assess our approach across various model types and datasets, we evaluate it with a speaker-independent deep clustering (DC) model and a speaker-dependent mask inference (MI) model. We report an improvement in separation performance of up to 1.5 dB in terms of source-to-distortion ratio (SDR) while maintaining an algorithmic latency of 8 ms.

AB - Time-frequency masking or spectrum prediction computed via short symmetric windows are commonly used in low-latency deep neural network (DNN) based source separation. In this paper, we propose the use of an asymmetric analysis-synthesis window pair which allows for training with targets with better frequency resolution, while retaining the low-latency during inference suitable for real-time speech enhancement or assisted hearing applications. In order to assess our approach across various model types and datasets, we evaluate it with a speaker-independent deep clustering (DC) model and a speaker-dependent mask inference (MI) model. We report an improvement in separation performance of up to 1.5 dB in terms of source-to-distortion ratio (SDR) while maintaining an algorithmic latency of 8 ms.

KW - Deep learning

KW - Training

KW - Time-frequency analysis

KW - Source separation

KW - Signal processing algorithms

KW - Europe

KW - Speech enhancement

KW - Monaural speaker separation

KW - Low latency

KW - Asymmetric windows

KW - Deep clustering

U2 - 10.23919/EUSIPCO54536.2021.9616165

DO - 10.23919/EUSIPCO54536.2021.9616165

M3 - Conference contribution

T3 - European Signal Processing Conference

SP - 301

EP - 305

BT - European Signal Processing Conference 2021

PB - IEEE

T2 - European Signal Processing Conference

Y2 - 23 August 2021 through 27 August 2021

ER -

Deep Neural Network Based Low-Latency Speech Separation with Asymmetric Analysis-Synthesis Window Pair

Abstract

Publication series

Conference

Keywords

Publication forum classification

Access to Document

Fingerprint

Cite this