From: Phone lattice reconstruction for embedded language recognition in LVCSR
Lattice reconstruction algorithm | Real time factor |
---|---|
A1. Time-aligned phone lattice | 0.19 |
A2. Phoneme-expanded 1-best lattice | 0.26 |
A3. Frame-expanded N-best lattice | 0.07 |
A3'. Pruned frame-expanded N-best lattice | 0.07 |
PRVSM, standard PRVSM system | 0.20 |
PRVSM-alt, PRVSM with LVCSR acoustic model | 0.44 |