---
_id: '61079'
abstract:
- lang: eng
text: "We propose a spatio-spectral, combined model-based and data-driven\r\ndiarization
pipeline consisting of TDOA-based segmentation followed by\r\nembedding-based
clustering. The proposed system requires neither access to\r\nmulti-channel training
data nor prior knowledge about the number or placement\r\nof microphones. It works
for both a compact microphone array and distributed\r\nmicrophones, with minor
adjustments. Due to its superior handling of\r\noverlapping speech during segmentation,
the proposed pipeline significantly\r\noutperforms the single-channel pyannote
approach, both in a scenario with a\r\ncompact microphone array and in a setup
with distributed microphones.\r\nAdditionally, we show that, unlike fully spatial
diarization pipelines, the\r\nproposed system can correctly track speakers when
they change positions."
author:
- first_name: Tobias
full_name: Cord-Landwehr, Tobias
id: '44393'
last_name: Cord-Landwehr
- first_name: Tobias
full_name: Gburrek, Tobias
id: '44006'
last_name: Gburrek
- first_name: Marc
full_name: Deegen, Marc
id: '70272'
last_name: Deegen
- first_name: Reinhold
full_name: Haeb-Umbach, Reinhold
id: '242'
last_name: Haeb-Umbach
citation:
ama: 'Cord-Landwehr T, Gburrek T, Deegen M, Haeb-Umbach R. Spatio-spectral diarization
of meetings by combining TDOA-based segmentation and speaker embedding-based
clustering. In: Proceedings of INTERSPEECH. ; 2025. doi:10.21437/Interspeech.2025-1663'
apa: Cord-Landwehr, T., Gburrek, T., Deegen, M., & Haeb-Umbach, R. (2025). Spatio-spectral
diarization of meetings by combining TDOA-based segmentation and speaker embedding-based
clustering. Proceedings of INTERSPEECH. Interspeech 2025, Rotterdam. https://doi.org/10.21437/Interspeech.2025-1663
bibtex: '@inproceedings{Cord-Landwehr_Gburrek_Deegen_Haeb-Umbach_2025, title={Spatio-spectral
diarization of meetings by combining TDOA-based segmentation and speaker embedding-based
clustering}, DOI={10.21437/Interspeech.2025-1663},
booktitle={Proceedings of INTERSPEECH}, author={Cord-Landwehr, Tobias and Gburrek,
Tobias and Deegen, Marc and Haeb-Umbach, Reinhold}, year={2025} }'
chicago: Cord-Landwehr, Tobias, Tobias Gburrek, Marc Deegen, and Reinhold Haeb-Umbach.
“Spatio-Spectral Diarization of Meetings by Combining TDOA-Based Segmentation
and Speaker Embedding-Based Clustering.” In Proceedings of INTERSPEECH,
2025. https://doi.org/10.21437/Interspeech.2025-1663.
ieee: 'T. Cord-Landwehr, T. Gburrek, M. Deegen, and R. Haeb-Umbach, “Spatio-spectral
diarization of meetings by combining TDOA-based segmentation and speaker embedding-based
clustering,” presented at the Interspeech 2025, Rotterdam, 2025, doi: 10.21437/Interspeech.2025-1663.'
mla: Cord-Landwehr, Tobias, et al. “Spatio-Spectral Diarization of Meetings by Combining
TDOA-Based Segmentation and Speaker Embedding-Based Clustering.” Proceedings
of INTERSPEECH, 2025, doi:10.21437/Interspeech.2025-1663.
short: 'T. Cord-Landwehr, T. Gburrek, M. Deegen, R. Haeb-Umbach, in: Proceedings
of INTERSPEECH, 2025.'
conference:
location: Rotterdam
name: Interspeech 2025
date_created: 2025-08-29T09:39:01Z
date_updated: 2025-11-10T09:06:47Z
ddc:
- '000'
department:
- _id: '54'
doi: 10.21437/Interspeech.2025-1663
external_id:
arxiv:
- '2506.16228'
file:
- access_level: open_access
content_type: application/pdf
creator: cord
date_created: 2025-08-29T09:43:32Z
date_updated: 2025-08-29T09:43:32Z
file_id: '61085'
file_name: main.pdf
file_size: 921918
relation: main_file
file_date_updated: 2025-08-29T09:43:32Z
has_accepted_license: '1'
language:
- iso: eng
oa: '1'
project:
- _id: '52'
name: Computing Resources Provided by the Paderborn Center for Parallel Computing
publication: Proceedings of INTERSPEECH
status: public
title: Spatio-spectral diarization of meetings by combining TDOA-based segmentation
and speaker embedding-based clustering
type: conference
user_id: '44393'
year: '2025'
...