Search papers, labs, and topics across Lattice.
2
0
3
SER models, often assumed to generalize well to synthesized speech, actually fail miserably, revealing their reliance on spurious correlations rather than genuine emotional understanding.
SLMs still lag behind omni language models in multi-turn conversational style control, as revealed by the new StyleBench benchmark.
