Skip to main content

Lund University Publications

LUND UNIVERSITY LIBRARIES

Alternative implementations of the Auxiliary Duplicating Permutation Invariant Training

Gulin, Jens LU orcid and Åström, Kalle LU orcid (2024)
Abstract
Simultaneous sound event localization and detection (SELD) for multi-source sound events is an open research field. The Multi-ACCDOA format is a popular way to handle activity-coupled sound events where the same class occurs at multiple locations at the same time. An important part is the Auxiliary Duplicating Permutation Invariant Training (ADPIT) paradigm that calculates the loss for order-agnosic output. The baseline system for the DCASE SELD challenge 2024 has an implementation of ADPIT. In this paper we discuss alternative ways to implement
ADPIT with the goal to reduce multiplications, to make the equivalent calculations faster. ADPIT duplicates output when there are fewer events than tracks. A brief discussion how this differs... (More)
Simultaneous sound event localization and detection (SELD) for multi-source sound events is an open research field. The Multi-ACCDOA format is a popular way to handle activity-coupled sound events where the same class occurs at multiple locations at the same time. An important part is the Auxiliary Duplicating Permutation Invariant Training (ADPIT) paradigm that calculates the loss for order-agnosic output. The baseline system for the DCASE SELD challenge 2024 has an implementation of ADPIT. In this paper we discuss alternative ways to implement
ADPIT with the goal to reduce multiplications, to make the equivalent calculations faster. ADPIT duplicates output when there are fewer events than tracks. A brief discussion how this differs from permutation invariant training without duplicated output is also included. The loss calculations are likely not the execution bottleneck in the current challenge setup, but ADPIT scales poorly for an increased number of tracks and improved efficiency is thus of general interest for audio localization. (Less)
Please use this url to cite or link to this publication:
author
and
organization
publishing date
type
Chapter in Book/Report/Conference proceeding
publication status
in press
subject
host publication
Proceedings of the Work-in-Progress Papers at the 14th International Conference on Indoor Positioning and Indoor Navigation (IPIN-WiP 2024)
language
English
LU publication?
yes
id
4e67bf36-e4b3-48b5-b1ff-c59772d0bc55
date added to LUP
2024-10-05 04:33:14
date last changed
2024-10-07 11:53:25
@inproceedings{4e67bf36-e4b3-48b5-b1ff-c59772d0bc55,
  abstract     = {{Simultaneous sound event localization and detection (SELD) for multi-source sound events is an open research field. The Multi-ACCDOA format is a popular way to handle activity-coupled sound events where the same class occurs at multiple locations at the same time. An important part is the Auxiliary Duplicating Permutation Invariant Training (ADPIT) paradigm that calculates the loss for order-agnosic output. The baseline system for the DCASE SELD challenge 2024 has an implementation of ADPIT. In this paper we discuss alternative ways to implement<br/>ADPIT with the goal to reduce multiplications, to make the equivalent calculations faster. ADPIT duplicates output when there are fewer events than tracks. A brief discussion how this differs from permutation invariant training without duplicated output is also included. The loss calculations are likely not the execution bottleneck in the current challenge setup, but ADPIT scales poorly for an increased number of tracks and improved efficiency is thus of general interest for audio localization.}},
  author       = {{Gulin, Jens and Åström, Kalle}},
  booktitle    = {{Proceedings of the Work-in-Progress Papers at the 14th International Conference on Indoor Positioning and Indoor Navigation (IPIN-WiP 2024)}},
  language     = {{eng}},
  title        = {{Alternative implementations of the Auxiliary Duplicating Permutation Invariant Training}},
  url          = {{https://lup.lub.lu.se/search/files/196591300/IPINwip_ARMPIT_final.pdf}},
  year         = {{2024}},
}