#include <julius/julius.h>
Go to the source code of this file.
Functions | |
static HTK_HMM_INFO * | initialize_HMM (JCONF_AM *amconf, Jconf *jconf) |
Read in an acoustic HMM from file and setup for recognition. | |
static HTK_HMM_INFO * | initialize_GSHMM (JCONF_AM *amconf) |
Initialize context-independent HMM for state selection with Gaussian Mixture Selection. | |
static HTK_HMM_INFO * | initialize_GMM (Jconf *jconf) |
Read and initialize an 1-state GMM for utterance verification and rejection. | |
static WORD_INFO * | initialize_dict (JCONF_LM *lmconf, HTK_HMM_INFO *hmminfo) |
Read in word dictionary from a file and setup for recognition. | |
static NGRAM_INFO * | initialize_ngram (JCONF_LM *lmconf, WORD_INFO *winfo) |
Read in word N-gram from file and setup for recognition. | |
boolean | j_load_am (Recog *recog, JCONF_AM *amconf) |
Load an acoustic model. | |
boolean | j_load_lm (Recog *recog, JCONF_LM *lmconf) |
Load a language model. | |
boolean | j_load_all (Recog *recog, Jconf *jconf) |
Read in all models for recognition. | |
static boolean | mfcc_config_is_same (JCONF_AM *amconf, MFCCCalc *mfcc) |
Check if parameter extraction configuration is the same between an AM configuration and a MFCC instance. | |
void | create_mfcc_calc_instances (Recog *recog) |
Create MFCC calculation instance for AM processing instances and GMM. | |
boolean | j_launch_recognition_instance (Recog *recog, JCONF_SEARCH *sconf) |
Launch a recognition process instance. | |
boolean | j_final_fusion (Recog *recog) |
Combine all loaded models and settings into one engine instance. |
These functions build everything needed for recognition: load models into memory, build data structures such as tree lexicon, and allocate work area for computation.
Definition in file m_fusion.c.
static HTK_HMM_INFO* initialize_HMM | ( | JCONF_AM * | amconf, | |
Jconf * | jconf | |||
) | [static] |
Read in an acoustic HMM from file and setup for recognition.
This functions reads HMM definitions from file, reads also a HMMList file, makes logical-to-physical model mapping, determine required parameter type, determine whether multi-path handling is needed, and find pause model in the definitions.
The feature vector extraction parameters are also finally determined in this function. Informations used for the determination is (1) the header values in hmmdefs, (2) embedded parameters in binary HMM if you are reading a binary HMM made with recent mkbinhmm, (3) user-specified parameters in jconf configurations (either by separatedly specified or by -htkconf options).
amconf | [in] AM configuration variables | |
jconf | [i/o] global configuration variables |
Definition at line 73 of file m_fusion.c.
Referenced by j_load_am().
static HTK_HMM_INFO* initialize_GSHMM | ( | JCONF_AM * | amconf | ) | [static] |
Initialize context-independent HMM for state selection with Gaussian Mixture Selection.
amconf | [in] AM configuratino variables |
Definition at line 166 of file m_fusion.c.
Referenced by j_load_am().
static HTK_HMM_INFO* initialize_GMM | ( | Jconf * | jconf | ) | [static] |
Read and initialize an 1-state GMM for utterance verification and rejection.
jconf | [in] global configuration variables |
Definition at line 198 of file m_fusion.c.
Referenced by j_load_all().
static WORD_INFO* initialize_dict | ( | JCONF_LM * | lmconf, | |
HTK_HMM_INFO * | hmminfo | |||
) | [static] |
Read in word dictionary from a file and setup for recognition.
Monophone-to-triphone conversion will be performed inside init_voca(). So, an HMM definition data that will be used with the LM should also be specified as an argument.
When reading dictionary for N-gram, sentence head silence word and tail silence word will be determined in this function. Also, when an option "-iwspword" is specified, this will insert a pause word at the last of the given dictionary.
lmconf | [in] LM configuration variables | |
hmminfo | [in] HMM definition of each phone in dictionary, for phone checking and monophone-to-triphone conversion. |
Definition at line 263 of file m_fusion.c.
Referenced by j_load_lm().
static NGRAM_INFO* initialize_ngram | ( | JCONF_LM * | lmconf, | |
WORD_INFO * | winfo | |||
) | [static] |
Read in word N-gram from file and setup for recognition.
When N-gram is specified in ARPA format, the behavior relies on whether N-grams are specified in "-nlr" and "-nrl". When either of them was solely specified, this function simply read it. If both are specified, it will read the RL model fully as a primary model, and additionally read only the 2-gram part or the LR model as the first pass LM.
Also, this function create mapping from dictionary words to LM entry.
lmconf | [in] LM configuration variables | |
winfo | [i/o] word dictionary that will be used with this N-gram. each word in the dictionary will be assigned to an N-gram entry here. |
Definition at line 353 of file m_fusion.c.
Referenced by j_load_lm().
Check if parameter extraction configuration is the same between an AM configuration and a MFCC instance.
amconf | [in] AM configuration parameters | |
mfcc | [in] MFCC calculation instance. |
Definition at line 709 of file m_fusion.c.
Referenced by create_mfcc_calc_instances().
void create_mfcc_calc_instances | ( | Recog * | recog | ) |
Create MFCC calculation instance for AM processing instances and GMM.
If more than one AM processing instance (or GMM) has the same configuration, the same MFCC calculation instance will be shared among them.
recog | [i/o] engine instance |
Definition at line 771 of file m_fusion.c.
Referenced by j_final_fusion().