Search papers, labs, and topics across Lattice.
1
0
3
23
Speech-aware LLMs are surprisingly bad at speaker verification, but a simple embedding injection trick closes the gap with dedicated systems while preserving the LLM's language abilities.
