Exploiting Parallel Audio Recordings to Enforce Device Invariance in CNN-based Acoustic Scene Classification

Paul Primus; Hamid Eghbal-Zadeh; David Eitelsebner; Khaled Koutini; Andreas Arzt; Gerhard Widmer

Exploiting Parallel Audio Recordings to Enforce Device Invariance in CNN-based Acoustic Scene Classification

Paul Primus
, Hamid Eghbal-Zadeh
, David Eitelsebner
, Khaled Koutini
, Andreas Arzt
, Gerhard Widmer

Institute of Computational Perception

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

Abstract

Distribution mismatches between the data seen at training and at application time remain a major challenge in all application areas of machine learning. We study this problem in the context of ma-chine listening (Task 1b of the DCASE 2019 Challenge). We pro-pose a novel approach to learn domain-invariant classifiers in an end-to-end fashion by enforcing equal hidden layer representations for domain-parallel samples, i.e. time-aligned recordings from different recording devices. No classification labels are needed for our domain adaptation (DA) method, which makes the data collection process cheaper. We show that our method improves the tar-get domain accuracy for both a toy dataset and an urban acoustic scenes dataset. We further compare our method to Maximum Mean Discrepancy-based DA and find it more robust to the choice of DA parameters. Our submission, based on this method, to DCASE 2019Task 1b gave us the 4th place in the team ranking.

Original language	English
Title of host publication	Proceedings of the Detection and Classification of Acoustic Scenes and Events 2019 Workshop (DCASE2019)
Number of pages	5
Publication status	Published - 2019

Fields of science

202002 Audiovisual media
102 Computer Sciences
102001 Artificial intelligence
102003 Image processing
102015 Information systems

JKU Focus areas

Digital Transformation

Cite this

@inproceedings{99609ecb45144c34b3381f06b5c42526,

title = "Exploiting Parallel Audio Recordings to Enforce Device Invariance in CNN-based Acoustic Scene Classification",

abstract = "Distribution mismatches between the data seen at training and at application time remain a major challenge in all application areas of machine learning. We study this problem in the context of ma-chine listening (Task 1b of the DCASE 2019 Challenge). We pro-pose a novel approach to learn domain-invariant classifiers in an end-to-end fashion by enforcing equal hidden layer representations for domain-parallel samples, i.e. time-aligned recordings from different recording devices. No classification labels are needed for our domain adaptation (DA) method, which makes the data collection process cheaper. We show that our method improves the tar-get domain accuracy for both a toy dataset and an urban acoustic scenes dataset. We further compare our method to Maximum Mean Discrepancy-based DA and find it more robust to the choice of DA parameters. Our submission, based on this method, to DCASE 2019Task 1b gave us the 4th place in the team ranking.",

author = "Paul Primus and Hamid Eghbal-Zadeh and David Eitelsebner and Khaled Koutini and Andreas Arzt and Gerhard Widmer",

year = "2019",

language = "English",

booktitle = "Proceedings of the Detection and Classification of Acoustic Scenes and Events 2019 Workshop (DCASE2019)",

}

Exploiting Parallel Audio Recordings to Enforce Device Invariance in CNN-based Acoustic Scene Classification. / Primus, Paul; Eghbal-Zadeh, Hamid; Eitelsebner, David et al.
Proceedings of the Detection and Classification of Acoustic Scenes and Events 2019 Workshop (DCASE2019). 2019.

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

TY - GEN

T1 - Exploiting Parallel Audio Recordings to Enforce Device Invariance in CNN-based Acoustic Scene Classification

AU - Primus, Paul

AU - Eghbal-Zadeh, Hamid

AU - Eitelsebner, David

AU - Koutini, Khaled

AU - Arzt, Andreas

AU - Widmer, Gerhard

PY - 2019

Y1 - 2019

N2 - Distribution mismatches between the data seen at training and at application time remain a major challenge in all application areas of machine learning. We study this problem in the context of ma-chine listening (Task 1b of the DCASE 2019 Challenge). We pro-pose a novel approach to learn domain-invariant classifiers in an end-to-end fashion by enforcing equal hidden layer representations for domain-parallel samples, i.e. time-aligned recordings from different recording devices. No classification labels are needed for our domain adaptation (DA) method, which makes the data collection process cheaper. We show that our method improves the tar-get domain accuracy for both a toy dataset and an urban acoustic scenes dataset. We further compare our method to Maximum Mean Discrepancy-based DA and find it more robust to the choice of DA parameters. Our submission, based on this method, to DCASE 2019Task 1b gave us the 4th place in the team ranking.

AB - Distribution mismatches between the data seen at training and at application time remain a major challenge in all application areas of machine learning. We study this problem in the context of ma-chine listening (Task 1b of the DCASE 2019 Challenge). We pro-pose a novel approach to learn domain-invariant classifiers in an end-to-end fashion by enforcing equal hidden layer representations for domain-parallel samples, i.e. time-aligned recordings from different recording devices. No classification labels are needed for our domain adaptation (DA) method, which makes the data collection process cheaper. We show that our method improves the tar-get domain accuracy for both a toy dataset and an urban acoustic scenes dataset. We further compare our method to Maximum Mean Discrepancy-based DA and find it more robust to the choice of DA parameters. Our submission, based on this method, to DCASE 2019Task 1b gave us the 4th place in the team ranking.

M3 - Conference proceedings

BT - Proceedings of the Detection and Classification of Acoustic Scenes and Events 2019 Workshop (DCASE2019)

ER -