Reproduce “Interpreting the Dimensions of Speaker Embedding Space” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Mark Huckvale
The authors test whether nine conventional acoustic descriptors can predict speaker embeddings from a Deep-Speaker model. On 9,836 Globe test speakers, an MLP using the descriptors predicts 256-dimensional embeddings at a mean cosine distance of 0.354, close to 0.349 for reconstruction using seven principal components, which explain 50.4% of variance. This compares embedding predictions, not verification or perceived voice similarity.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.