How LSTM is Integrated in the Functional Link Adaptive Filter

Alireza Nezamdoust; Danilo Comminiello; Amir  Hussain; Kia Dashtipour; Mandar Gogate

doi:10.21437/AVSEC.2025-15

How LSTM is Integrated in the Functional Link Adaptive Filter

Alireza Nezamdoust^*, Danilo Comminiello, Amir Hussain, Kia Dashtipour, Mandar Gogate

^*Corresponding author for this work

Institute of Signal Processing

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

Abstract

The SFLAF with LSTM model is a hybrid nonlinear acoustic echo cancellation (NAEC) system designed to remove echo from a microphone signal d[n] using the far-end signal x[n]. It combines the Split Functional Link Adaptive Filter (SFLAF), which handles both linear and nonlinear echo components, with a Long Short-Term Memory (LSTM) network, which refines the residual error as a cascaded post-processing stage. Below, I will describe the architecture, its cascaded nature, and the mathematical formulation.

Original language	English
Title of host publication	Proceedings of the AVSEC 2025, 26th edition of the Interspeech Conference
Subtitle of host publication	4th Cogmhear Audio-Visual Speech Enhancement Challenge (AVSE)
Pages	67-71
Number of pages	5
DOIs	https://doi.org/10.21437/AVSEC.2025-15
Publication status	Published - 2025

Fields of science

202034 Control engineering
202017 Embedded systems
202015 Electronics
202030 Communication engineering
202028 Microelectronics
202027 Mechatronics
102019 Machine learning
202040 Transmission technology
202025 Power electronics
202041 Computer engineering
202037 Signal processing
202023 Integrated circuits
202036 Sensor systems
202022 Information technology
102 Computer Sciences
202 Electrical Engineering, Electronics, Information Engineering

JKU Focus areas

Digital Transformation

Access to Document

10.21437/AVSEC.2025-15

https://www.isca-archive.org/avsec_2025/nezamdoust25_avsec.pdf

LCM - Center for Symbiotic Mechatronics - Förderphase 2
Amrhein, W. (Researcher), Baumgartner, W. (Researcher), Buckwar, E. (Researcher), Egyed, A. (Researcher), Gattringer, R. (Researcher), Gerardo-Giorda, L. (Researcher), Gittler, P. (Researcher), Huemer, M. (Researcher), Jakoby, B. (Researcher), Krommer, M. (Researcher), Küng, J. (Researcher), Müller, A. (Researcher), Pirker, S. (Researcher), Saminger-Platz, S. (Researcher), Schagerl, M. (Researcher), Scheidl, R. (Researcher), Schlacher, K. (Researcher), Schöberl, M. (Researcher), Springer, A. (Researcher), Widmer, G. (Researcher), Zagar, B. (Researcher), Zeman, K. (Researcher), Zoitl, A. (Researcher) & Hoffelner, J. (PI)
01.01.2022 → 31.12.2026
Project: Funded research › FFG - Austrian Research Promotion Agency

Cite this

@inproceedings{b66daae6ab5f4838a75f476fdd844ef0,

title = "How LSTM is Integrated in the Functional Link Adaptive Filter",

abstract = "The SFLAF with LSTM model is a hybrid nonlinear acoustic echo cancellation (NAEC) system designed to remove echo from a microphone signal d[n] using the far-end signal x[n]. It combines the Split Functional Link Adaptive Filter (SFLAF), which handles both linear and nonlinear echo components, with a Long Short-Term Memory (LSTM) network, which refines the residual error as a cascaded post-processing stage. Below, I will describe the architecture, its cascaded nature, and the mathematical formulation.",

author = "Alireza Nezamdoust and Danilo Comminiello and Amir Hussain and Kia Dashtipour and Mandar Gogate",

year = "2025",

doi = "10.21437/AVSEC.2025-15",

language = "English",

pages = "67--71",

booktitle = "Proceedings of the AVSEC 2025, 26th edition of the Interspeech Conference",

}

How LSTM is Integrated in the Functional Link Adaptive Filter. / Nezamdoust, Alireza; Comminiello, Danilo; Hussain, Amir et al.
Proceedings of the AVSEC 2025, 26th edition of the Interspeech Conference: 4th Cogmhear Audio-Visual Speech Enhancement Challenge (AVSE). 2025. p. 67-71.

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

TY - GEN

T1 - How LSTM is Integrated in the Functional Link Adaptive Filter

AU - Nezamdoust, Alireza

AU - Comminiello, Danilo

AU - Hussain, Amir

AU - Dashtipour, Kia

AU - Gogate, Mandar

PY - 2025

Y1 - 2025

N2 - The SFLAF with LSTM model is a hybrid nonlinear acoustic echo cancellation (NAEC) system designed to remove echo from a microphone signal d[n] using the far-end signal x[n]. It combines the Split Functional Link Adaptive Filter (SFLAF), which handles both linear and nonlinear echo components, with a Long Short-Term Memory (LSTM) network, which refines the residual error as a cascaded post-processing stage. Below, I will describe the architecture, its cascaded nature, and the mathematical formulation.

AB - The SFLAF with LSTM model is a hybrid nonlinear acoustic echo cancellation (NAEC) system designed to remove echo from a microphone signal d[n] using the far-end signal x[n]. It combines the Split Functional Link Adaptive Filter (SFLAF), which handles both linear and nonlinear echo components, with a Long Short-Term Memory (LSTM) network, which refines the residual error as a cascaded post-processing stage. Below, I will describe the architecture, its cascaded nature, and the mathematical formulation.

U2 - 10.21437/AVSEC.2025-15

DO - 10.21437/AVSEC.2025-15

M3 - Conference proceedings

SP - 67

EP - 71

BT - Proceedings of the AVSEC 2025, 26th edition of the Interspeech Conference

ER -

How LSTM is Integrated in the Functional Link Adaptive Filter

Abstract

Fields of science

JKU Focus areas

Access to Document

Projects

LCM - Center for Symbiotic Mechatronics - Förderphase 2

Cite this