Publication Details
Joint Training of Speaker Embedding Extractor, Speech and Overlap Detection for Diarization
Landini Federico Nicolás (DCGM FIT BUT)
Klement Dominik, Ing. (DCGM FIT BUT)
Diez Sánchez Mireia, M.Sc., Ph.D. (DCGM FIT BUT)
Silnova Anna, MSc., Ph.D. (DCGM FIT BUT)
Delcroix Marc (NTT)
Burget Lukáš, doc. Ing., Ph.D. (DCGM FIT BUT)
speaker diarization, speaker embedding, voice activity detection, overlapped speech detection
In spite of the popularity of end-to-end diarization systems nowadays, modular systems comprised of voice activity detection (VAD), speaker embedding extraction plus clustering, and overlapped speech detection (OSD) plus handling still attain competitive performance in many conditions. However, one of the main drawbacks of modular systems is the need to run (and train) different modules independently. In this work, we propose an approach to jointly train a model to produce speaker embeddings, VAD and OSD simultaneously and reach competitive performance at a fraction of the inference time of a modular approach. Furthermore, the joint inference leads to a simplified overall pipeline which brings us one step closer to a unified clustering-based method that can be trained end-to-end towards a diarization-specific objective.
@INPROCEEDINGS{FITPUB13567,
author = "Petr P\'{a}lka and Nicol\'{a}s Federico Landini and Dominik Klement and Mireia S\'{a}nchez Diez and Anna Silnova and Marc Delcroix and Luk\'{a}\v{s} Burget",
title = "Joint Training of Speaker Embedding Extractor, Speech and Overlap Detection for Diarization",
pages = "1--5",
booktitle = "Proceedings of Eusipco 2025",
year = 2025,
location = "Palermo, IT",
publisher = "IEEE Signal Processing Society",
language = "english",
url = "https://www.fit.vut.cz/research/publication/13567"
}