|
Julius 4.2
|
LM cache for the 1st pass. [詳細]
#include <wchmm.h>
変数 | |
| LOGPROB * | probcache |
| Word-internal factoring cache indexed by scid, holding last score. | |
| WORD_ID * | lastwcache |
| Word-internal factoring cache indexed by scid, holding last N-gram entry ID. | |
| LOGPROB ** | iw_sc_cache |
| Cross-word factoring cache to hold last-word-dependent factoring score toward word head nodes. | |
| int | iw_cache_num |
| Maximum size of cross-word factoring cache iw_sc_cache per last word. | |
Cross-word factoring cache to hold last-word-dependent factoring score toward word head nodes.
Cached values will be stored as [last_nword][n], where n is the number of word-head node on which the last_nword-dependent N-gram factoring value should be computed on cross-word transition. In 1-gram factoring, n equals to wchmm->isolatenum, the number of isolated (not shared) word-head nodes. In 2-gram factoring, n simply equals to wchmm->startnum, the number of all word-head nodes.
The cache area will be allocated per the previous word when they appeared while search. It will retain across the speech stream, so the cache area will grow to an extent as recognition was done for many files.
参照元 max_successor_cache_free(), max_successor_cache_init(), max_successor_prob_iw(), と max_successor_prob_iw_free().
Maximum size of cross-word factoring cache iw_sc_cache per last word.
The value is set in max_successor_cache_init().
参照元 max_successor_cache_init(), max_successor_prob_iw(), と max_successor_prob_iw_free().
1.7.4