Benchmark
VoxPopuli
A multilingual speech dataset and benchmark commonly used for evaluation of tasks such as speech recognition and clustering of speaker attributes such as gender.
Timeline (last 90 days)
Oct 28
Probing layer 2 (mean + standard deviation pooling) of EmbeddingGemma-2 raises VoxPopuli gender clustering from 0.05% at the final output to 66.6% (layer 3)/above 65% (layer 2), and IEMOCAP gender classification from 66.4% to 83.0%.
Unverified50%