Skip to main content

Table 4 Performance of parallel frontends

From: Phone lattice reconstruction for embedded language recognition in LVCSR

Lattice reconstruction algorithm EER (%)
A0. Word-based phone lattice 21.28
A1. Time-aligned phone lattice 15.89
A2. Phoneme-expanded 1-best lattice 15.79
A3. Frame-expanded N-best lattice 15.40
A3'. Pruned frame-expanded N-best lattice 14.98
PRVSM, standard PRVSM system 14.55
PRVSM-alt, PRVSM with LVCSR acoustic model 14.97
  1. EERs of the A3' system were obtained by using the 84% pruning beam width for Mandarin and 81% for English