Search papers, labs, and topics across Lattice.
1
0
3
Ditch fixed chunk sizes: TC-BiMamba unlocks faster, more memory-efficient training for bidirectional Mamba ASR models, enabling unified streaming and non-streaming decoding.
