#include <vithist.h>
Public Attributes | |
| vithist_entry_t ** | entry |
| int32 * | frame_start |
| int32 | n_entry |
| int32 | n_frm |
| int32 | n_ci |
| int32 | bghist |
| int32 | wbeam |
| int32 * | bestscore |
| int32 * | bestvh |
| vh_lms2vh_t ** | lms2vh_root |
| glist_t | lwidlist |
Memory management of Viterbi history entries done in blocks. Initially, one block of VITHIST_BLKSIZE entries allocated. If exhausted, another block allocated, and so on. So we can have several discontiguous blocks allocated. Entries are identified by a global, running sequence no.
| int32* vithist_t::bestscore |
Best word exit score in each frame
| int32* vithist_t::bestvh |
Vithist entry ID with the best exit score in each frame
| int32 vithist_t::bghist |
If TRUE (bigram-mode) only one entry/word/frame; otherwise multiple entries allowed, one per distinct LM state
entry[i][j]= j-th entry in the i-th block allocated
| int32* vithist_t::frame_start |
For each frame, the first vithist ID in that frame; (the last is just before the first of the next frame)
lms2vh[w]= Root of LM states ending in w in current frame
| glist_t vithist_t::lwidlist |
List of LM word IDs with entries in lms2vh_root
| int32 vithist_t::n_ci |
No. of CI phones
| int32 vithist_t::n_entry |
Total #entries used (generates global seq no. or ID)
| int32 vithist_t::n_frm |
No. of frames processed so far in this utterance
| int32 vithist_t::wbeam |
Pruning beamwidth
1.6.1