Pretraining Multi-Speaker Identification for Neural Speaker Diarization
Fuente:
arXiv
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866912404182401024 |
|---|---|
| author | Horiguchi, Shota Ando, Atsushi Delcroix, Marc Tawara, Naohiro |
| author_facet | Horiguchi, Shota Ando, Atsushi Delcroix, Marc Tawara, Naohiro |
| contents | End-to-end speaker diarization enables accurate overlap-aware diarization by jointly estimating multiple speakers' speech activities in parallel. This approach is data-hungry, requiring a large amount of labeled conversational data, which cannot be fully obtained from real datasets alone. To address this issue, large-scale simulated data is often used for pretraining, but it requires enormous storage and I/O capacity, and simulating data that closely resembles real conversations remains challenging. In this paper, we propose pretraining a model to identify multiple speakers from an input fully overlapped mixture as an alternative to pretraining a diarization model. This method eliminates the need to prepare a large-scale simulated dataset while leveraging large-scale speaker recognition datasets for training. Through comprehensive experiments, we demonstrate that the proposed method enables a highly accurate yet lightweight local diarization model without simulated conversational data. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2505_24545 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Pretraining Multi-Speaker Identification for Neural Speaker Diarization Horiguchi, Shota Ando, Atsushi Delcroix, Marc Tawara, Naohiro Audio and Speech Processing Sound End-to-end speaker diarization enables accurate overlap-aware diarization by jointly estimating multiple speakers' speech activities in parallel. This approach is data-hungry, requiring a large amount of labeled conversational data, which cannot be fully obtained from real datasets alone. To address this issue, large-scale simulated data is often used for pretraining, but it requires enormous storage and I/O capacity, and simulating data that closely resembles real conversations remains challenging. In this paper, we propose pretraining a model to identify multiple speakers from an input fully overlapped mixture as an alternative to pretraining a diarization model. This method eliminates the need to prepare a large-scale simulated dataset while leveraging large-scale speaker recognition datasets for training. Through comprehensive experiments, we demonstrate that the proposed method enables a highly accurate yet lightweight local diarization model without simulated conversational data. |
| title | Pretraining Multi-Speaker Identification for Neural Speaker Diarization |
| topic | Audio and Speech Processing Sound |
| url | https://arxiv.org/abs/2505.24545 |