Dynamic nsNET2: Efficient Deep Noise Suppression with Early Exiting

Riccardo Miccini; Alaa Zniber; Clément Laroche; Tobias Piechowiak; Martin Schoeberl; Luca Pezzarossa; Ouassim Karrakchou; Jens Sparsø; Mounir Ghogho

doi:10.1109/MLSP55844.2023.10285925

Dynamic nsNET2: Efficient Deep Noise Suppression with Early Exiting

Riccardo Miccini, Alaa Zniber, Clément Laroche, Tobias Piechowiak, Martin Schoeberl, Luca Pezzarossa, Ouassim Karrakchou, Jens Sparsø, Mounir Ghogho

Research output: Chapter in Book/Report/Conference proceeding › Article in proceedings › Research › peer-review

72 Downloads (Pure)

Abstract

Although deep learning has made strides in the field of deep noise suppression, leveraging deep architectures on resourceconstrained devices still proved challenging. Therefore, we present an early-exiting model based on nsNet2 that provides several levels of accuracy and resource savings by halting computations at different stages. Moreover, we adapt the original architecture by splitting the information flow to take into account the injected dynamism. We show the trade-offs between performance and computational complexity based on established metrics.

Original language	English
Title of host publication	Proceedings of the 2023 IEEE 33rd International Workshop on Machine Learning for Signal Processing (MLSP)
Number of pages	6
Publisher	IEEE
Publication date	2023
ISBN (Print)	979-8-3503-2412-9
ISBN (Electronic)	979-8-3503-2411-2
DOIs	https://doi.org/10.1109/MLSP55844.2023.10285925
Publication status	Published - 2023
Event	2023 IEEE 33rd International Workshop on Machine Learning for Signal Processing - Rome, Italy, Rome, Italy Duration: 17 Sept 2023 → 20 Sept 2023

Conference

Conference	2023 IEEE 33rd International Workshop on Machine Learning for Signal Processing
Location	Rome, Italy
Country/Territory	Italy
City	Rome
Period	17/09/2023 → 20/09/2023

Keywords

Deep Noise Suppression
Dynamic Neural Networks
Early-exiting

Access to Document

10.1109/MLSP55844.2023.10285925

FulltextAccepted author manuscript, 481 KB

OpenUrl availability

Full text

Cite this

@inproceedings{b99b8563d40f46e38cc06b6d5ad3f9d4,

title = "Dynamic nsNET2: Efficient Deep Noise Suppression with Early Exiting",

abstract = "Although deep learning has made strides in the field of deep noise suppression, leveraging deep architectures on resourceconstrained devices still proved challenging. Therefore, we present an early-exiting model based on nsNet2 that provides several levels of accuracy and resource savings by halting computations at different stages. Moreover, we adapt the original architecture by splitting the information flow to take into account the injected dynamism. We show the trade-offs between performance and computational complexity based on established metrics.",

keywords = "Deep Noise Suppression, Dynamic Neural Networks, Early-exiting",

author = "Riccardo Miccini and Alaa Zniber and Cl{\'e}ment Laroche and Tobias Piechowiak and Martin Schoeberl and Luca Pezzarossa and Ouassim Karrakchou and Jens Spars{\o} and Mounir Ghogho",

year = "2023",

doi = "10.1109/MLSP55844.2023.10285925",

language = "English",

isbn = "979-8-3503-2412-9",

booktitle = "Proceedings of the 2023 IEEE 33rd International Workshop on Machine Learning for Signal Processing (MLSP)",

publisher = "IEEE",

address = "United States",

note = "2023 IEEE 33rd International Workshop on Machine Learning for Signal Processing , MLSP ; Conference date: 17-09-2023 Through 20-09-2023",

}

Miccini, R, Zniber, A, Laroche, C, Piechowiak, T, Schoeberl, M , Pezzarossa, L, Karrakchou, O, Sparsø, J & Ghogho, M 2023, Dynamic nsNET2: Efficient Deep Noise Suppression with Early Exiting. in Proceedings of the 2023 IEEE 33rd International Workshop on Machine Learning for Signal Processing (MLSP). IEEE, 2023 IEEE 33rd International Workshop on Machine Learning for Signal Processing , Rome, Italy, 17/09/2023. https://doi.org/10.1109/MLSP55844.2023.10285925

TY - GEN

T1 - Dynamic nsNET2

T2 - 2023 IEEE 33rd International Workshop on Machine Learning for Signal Processing

AU - Miccini, Riccardo

AU - Zniber, Alaa

AU - Laroche, Clément

AU - Piechowiak, Tobias

AU - Schoeberl, Martin

AU - Pezzarossa, Luca

AU - Karrakchou, Ouassim

AU - Sparsø, Jens

AU - Ghogho, Mounir

PY - 2023

Y1 - 2023

N2 - Although deep learning has made strides in the field of deep noise suppression, leveraging deep architectures on resourceconstrained devices still proved challenging. Therefore, we present an early-exiting model based on nsNet2 that provides several levels of accuracy and resource savings by halting computations at different stages. Moreover, we adapt the original architecture by splitting the information flow to take into account the injected dynamism. We show the trade-offs between performance and computational complexity based on established metrics.

AB - Although deep learning has made strides in the field of deep noise suppression, leveraging deep architectures on resourceconstrained devices still proved challenging. Therefore, we present an early-exiting model based on nsNet2 that provides several levels of accuracy and resource savings by halting computations at different stages. Moreover, we adapt the original architecture by splitting the information flow to take into account the injected dynamism. We show the trade-offs between performance and computational complexity based on established metrics.

KW - Deep Noise Suppression

KW - Dynamic Neural Networks

KW - Early-exiting

U2 - 10.1109/MLSP55844.2023.10285925

DO - 10.1109/MLSP55844.2023.10285925

M3 - Article in proceedings

SN - 979-8-3503-2412-9

BT - Proceedings of the 2023 IEEE 33rd International Workshop on Machine Learning for Signal Processing (MLSP)

PB - IEEE

Y2 - 17 September 2023 through 20 September 2023

ER -

Dynamic nsNET2: Efficient Deep Noise Suppression with Early Exiting

Abstract

Conference

Keywords

Access to Document

OpenUrl availability

Fingerprint

Cite this