CREMA-D
A benchmark dataset for speech emotion recognition containing emotional speech samples from multiple speakers, commonly used to evaluate the emotion classification capability of audio models.
Timeline (last 90 days)
Existing LLM-driven Omni embedding models score 95-99% on multilingual text-to-audio retrieval on the MAEB leaderboard, but only 1%-6% on unsupervised audio clustering and about 20% on speech emotion recognition (CREMA-D).
Probing layer 6 (mean pooling) of EmbeddingGemma-2 doubles CREMA-D emotion classification from 19.7% to 39.5%.
All Facts (2)
Existing LLM-driven Omni embedding models score 95-99% on multilingual text-to-audio retrieval on the MAEB leaderboard, but only 1%-6% on unsupervised audio clustering and about 20% on speech emotion recognition (CREMA-D).
50%UnverifiedProbing layer 6 (mean pooling) of EmbeddingGemma-2 doubles CREMA-D emotion classification from 19.7% to 39.5%.
50%