Emotion is a middle-layer property
A linear probe on frozen features, across all 25 hidden states, six encoders and seven corpora. Accuracy climbs to layer 9 and then falls away. The last layer is the worst place to look, which is exactly where a classifier head would normally sit.
The encoder barely matters. The corpus does.
Six encoders separate by under four points once averaged across the grid. The corpora they are measured on span thirty-four. Most of what a benchmark number tells you is which recordings it was computed on.
Language is the real barrier
Pool four languages and every encoder looks excellent. Hold one out and the answer depends entirely on which one, and adding a language from another family collapses it.
Some encoders read the transcript
EMIS pairs audio with words whose emotion deliberately contradicts the delivery. A model hearing the voice should ignore them. Three of six do not.
What a probe reads is not what the model has
Sparse autoencoders rewrite each activation as 8192 features with 32 active at a time. 394 look like clean emotion features. Because CREMA-D has every actor speak the same twelve sentences in six emotions, we can ask whether a feature says more about the emotion than about the sentence or the speaker. Most cannot, and 61 remain.
What did not survive
Fine-tuning lost to frozen probing in 29 of 30 LoRA runs. Adversarial and CCA debiasing both cost more accuracy than the bias they removed. The IEMOCAP ceiling near 0.68 turned out to be annotator disagreement rather than model capacity: restricting to clips the humans agreed on lifts accuracy by up to 0.19. And the 61 surviving sparse features are shown by sufficiency rather than necessity, since ablating them does not reduce accuracy.