Search papers, labs, and topics across Lattice.
1
0
3
8
LLM-based ASR can be sped up by 4.4x with minimal accuracy loss by using a CTC encoder to speculatively generate draft transcriptions.
