59 * 1, Important Bug fixes, a local pointer is used when realloc is needed. This causes invalid writing of the memory, 2, Acoustic scores of the last segment in IBM lattice generation couldn't be found in the past. Now, this could be handled by the optional acoustic scores in the node of lattice. Application code is not yet checked-in
62 * 3 changes. 1, Code was added for full triphone implementation, not yet working. 2, Senone scale is removed from vithist table. This was a bug introduced during some fixes in CALO.
65 * vithist.c is the true source of memory leaks in the past for full cwtp expansion. There are two changes made to avoid this happen, 1, instead of using ve->rc_info as the indicator whether something should be done, used a flag bFullExpand to control it. 2, avoid doing direct C-struct copy (like *ve = *tve), it becomes the reason of why memory are leaked and why the code goes wrong.
74 * 1, Added arguments for turning on/off LTS rules. 2, Added arguments for turning on/off composite triphones. 3, Moved dict2pid deallocation back to dict2pid. 4, Tidying up the clean up code.
77 * All lattice-related functions are not incorporated into vithist. So-called "lattice" is essentially the predecessor of vithist_t and fsg_history_t. Later when vithist_t support by right context score and history. It should replace both of them.
86 * 1, Added reporting flag for vithist_init. 2, Added a flag to allow using words other than silence to be the last word for backtracing. 3, Fixed doxygen documentation. 4, Add keyword.
87 *
88 * Revision 1.10 2005/06/16 04:59:10 archan
89 * Sphinx3 to s3.generic, a gentle-refactored version of Dave's change in senone scale.
90 *
91 * Revision 1.9 2005/06/13 04:02:59 archan
92 * Fixed most doxygen-style documentation under libs3decoder.
93 *
94 * Revision 1.8 2005/05/26 00:46:59 archan
95 * Added functionalities that such that <sil> will not be inserted at the end of the utterance.
96 *
97 * Revision 1.7 2005/04/25 23:53:35 archan
98 * 1, Some minor modification of vithist_t, vithist_rescore can now support optional LM rescoring, vithist also has its own reporting routine. A new argument -lmrescore is also added in decode and livepretend. This can switch on and off the rescoring procedure. 2, I am reaching the final difficulty of mode 5 implementation. That is, to implement an algorithm which dynamically decide which tree copies should be entered. However, stuffs like score propagation in the leave nodes and non-leaves nodes are already done. 3, As briefly mentioned in 2, implementation of rescoring , which used to happened at leave nodes are now separated. The current implementation is not the most clever one. Wish I have time to change it before check-in to the canonical.
99 *
100 * Revision 1.6 2005/04/21 23:50:26 archan
101 * Some more refactoring on the how reporting of structures inside kbcore_t is done, it is now 50% nice. Also added class-based LM test case into test-decode.sh.in. At this moment, everything in search mode 5 is already done. It is time to test the idea whether the search can really be used.
102 *
103 * Revision 1.5 2005/04/20 03:46:30 archan
104 * factor dag header writer into vithist.[ch], do the corresponding change for lm_t