ISCA Archive Odyssey 2022
ISCA Archive Odyssey 2022

Magnitude-Aware Probabilistic Speaker Embeddings

Nikita Kuzmin, Igor Fedorov, Alexey Sholokhov

Recently, hyperspherical embeddings have established themselves as a dominant technique for face and voice recognition. Specifically, Euclidean space vector embeddings are learned to encode person-specific information in their direction while ignoring the magnitude. However, recent studies have shown that the magnitudes of the embeddings extracted by deep neural networks may indicate the quality of the corresponding inputs. This paper explores the properties of the magnitudes of the embeddings related to quality assessment and out-of-distribution detection. We propose a new probabilistic speaker embedding extractor using the information encoded in the embedding magnitude and leverage it in the speaker verification pipeline. We also propose several quality-aware diarization methods and incorporate the magnitudes in those. Our results indicate significant improvements over magnitude-agnostic baselines both in speaker verification and diarization tasks.

doi: 10.21437/Odyssey.2022-1

Cite as: Kuzmin, N., Fedorov, I., Sholokhov, A. (2022) Magnitude-Aware Probabilistic Speaker Embeddings. Proc. The Speaker and Language Recognition Workshop (Odyssey 2022), 1-8, doi: 10.21437/Odyssey.2022-1

  author={Nikita Kuzmin and Igor Fedorov and Alexey Sholokhov},
  title={{Magnitude-Aware Probabilistic Speaker Embeddings}},
  booktitle={Proc. The Speaker and Language Recognition Workshop (Odyssey 2022)},