The input utterance to decode has 668 frames: U (U_feats.txt)
The decoding graph: HCLG (HCLG.pdf)
U o HCLG:
Decode(): creates the active_toks_ list for every frame (decode.pdf).
GetRawLattice: creates the state-level lattice based on the active_toks_ list (rawlat.pdf).
Determinize(): determinize the raw lattice (detlat.pdf).