This repository contains a ResNet34-based d-vector system from [1] to extract speaker embeddings
trained on VoxCeleb augmented with MUSAN and simulated room impulse responses.
It was used for [2] and [3]. It is used in [4], which contains the code for [3].
See [4], to get an example, how you can use this speaker embedding extractor.
- [1] T. Cord-Landwehr, C. Boeddeker, C. Zorila, R. Doddipatla, and ˘
R. Haeb-Umbach, “Frame-wise and overlap-robust speaker embeddings for meeting diarization,” in IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2023,
pp. 1–5.
- [2] C. Boeddeker, T. Cord-Landwehr, T. von Neumann, and R. Haeb-
Umbach, “Multi-stage diarization refinement for the CHiME-7
DASR scenario,” in Proc. CHiME, 2023, pp. 51–56.
- [3] C. Boeddeker, T. Cord-Landwehr, and R. Haeb-
Umbach, “Once more Diarization: Improving meeting transcription systems through
segment-level speaker reassignment,” accepted for ISCA Interspeech,
