49 * Added interface for 1, doing multi stream gmm computation with/without composite senone. 2, doing gmm computation (ms or ss optimized) with/wihout composite senone. Haven't tested on the SCHMM on s3.x yet. I think it will work though.
61 * A wrapper that provide the function pointer interface of approx_cont_mgau_ci_eval and approx_cont_mgau_frame_eval. They are used in srch_gmm_compute_lv1 and srch_gmm_compute_lv2 respectively. This will also be the home of other gmm computation routine. (Say the s3.0 version of GMM computation)
62 *
63 * Revision 1.2 2005/06/13 04:02:59 archan
64 * Fixed most doxygen-style documentation under libs3decoder.
65 *
66 * Revision 1.1 2005/04/22 04:22:37 archan
67 * Add gmm_wrap, this will share code across op_mode 4 and op_mode 5. Also it also separate active senone selection into a different process. I hope this is the final step before making the WST search works. At the current stage, the code of mode-5 looks very much alike mode-4. This is intended because in Prototype 4, tail sharing will be used to reduce memory.