libTLK  1.3.1
Defines | Functions
Viterbi training of HMMs
Algorithms

Defines

#define tL_train_viterbi_hmm(HVTRAINER, COUNTS, C, FEA)   tL_train_viterbi_hmm_seg ( HVTRAINER, COUNTS, C, FEA, 0, (FEA)->cols )
 Trains an HMM from a sample using the Viterbi algorithm.

Functions

 tL_hmm_vtrainer_free (tLHMMVTrainer *hvtrainer)
 Frees memory.
 tL_hmm_vtrainer_new (const tLAModel *amodel, char **err)
 Gets a tLHMMVTrainer.
 tL_train_viterbi_hmm_seg (tLHMMVTrainer *hvtrainer, tLAMiCounts *counts, const int c, const tLFea *fea, const int begin, const int nframes)
 Trains an HMM from a sample segment using the Viterbi algorithm.

Define Documentation

#define tL_train_viterbi_hmm (   HVTRAINER,
  COUNTS,
  C,
  FEA 
)    tL_train_viterbi_hmm_seg ( HVTRAINER, COUNTS, C, FEA, 0, (FEA)->cols )

Trains an HMM from a sample using the Viterbi algorithm.

This macro is equivalent to tL_train_viterbi_hmm_seg but using all frames contained in fea.

Parameters:
HVTRAINERThe HMM Viterbi trainer.
COUNTSInteger counts related to AMODEL.
CThe index of the HMM to be trained.
FEAA feature vector manager, which must contain a sequence of feature vectors of an adequate type for the AMODEL.
Returns:
The log-probability of the maximum probability path.

Function Documentation

tL_hmm_vtrainer_free ( tLHMMVTrainer *  hvtrainer)

Frees memory.

Frees the memory allocated for the HMM Viterbi trainer.

Parameters:
hvtrainerThe HMM Viterbi trainer.
tL_hmm_vtrainer_new ( const tLAModel amodel,
char **  err 
)

Gets a tLHMMVTrainer.

This function creates an HMM Viterbi trainer used to train HMMs with the Viterbi algorithm. Currently supported types are: See tLHMMInitializer.

Parameters:
amodelThe acoustic model.
errPointer to string variable. If not NULL, an error message is allocated in the variable in case of error.
Returns:
The HMM Viterbi trainer, or NULL in case of error.
tL_train_viterbi_hmm_seg ( tLHMMVTrainer *  hvtrainer,
tLAMiCounts counts,
const int  c,
const tLFea fea,
const int  begin,
const int  nframes 
)

Trains an HMM from a sample segment using the Viterbi algorithm.

This function finds the maximum probability path for the specified HMM of input frames, which are a segment of the given feature vector sequence, and for each transition used it increases the related count by one. Furthermore, the maximum probability path is used to assign the input frames to states, then, states counts are increased as in tL_init_hmm.

Parameters:
hvtrainerThe HMM Viterbi trainer.
countsInteger counts related to amodel.
cThe index of the HMM to be trained.
feaA feature vector manager, which must contain a sequence of feature vectors of an adequate type for the amodel.
beginThe first input frame.
nframesThe number of input frames.
Returns:
The log-probability of the maximum probability path.
 All Data Structures Variables