libTLK  1.3.1
Functions
EM training from transcriptions of acoustic models
Algorithms

Functions

 tL_trainer_free (tLTrainer *trainer)
 Frees memory.
 tL_trainer_new (tLAModel *amodel, tLAMlogCounts *counts, const tLSeqs *seqs, tLFloat **occs, char **err)
 Gets a tLTrainer.
 tL_train (tLTrainer *trainer, const tLFea *fea, const size_t seq)
 Updates an acoustic model's counts according to the EM algorithm.

Function Documentation

tL_train ( tLTrainer *  trainer,
const tLFea fea,
const size_t  seq 
)

Updates an acoustic model's counts according to the EM algorithm.

This function increases the acoustic model's counts, which are stored in the provided trainer, from one sample (feature vector sequence) and its transcription, using the EM algorithm in order to obtain the logarithmic counts. The sequence is specified as an integer code according to the sequence manager used during the creation of the trainer.

Parameters:
trainerThe trainer.
feaA feature vector manager, which must contain a sequence of feature vectors of an adequate type for the amodel.
seqThe code of the transcription of the sample.
Returns:
The sample probability for that transcription; -HUGE_VAL in case the sample cannot be used to update the counts.
tL_trainer_free ( tLTrainer *  trainer)

Frees memory.

Frees the memory allocated for the trainer.

Parameters:
trainerThe trainer.
tL_trainer_new ( tLAModel amodel,
tLAMlogCounts counts,
const tLSeqs seqs,
tLFloat **  occs,
char **  err 
)

Gets a tLTrainer.

This function creates a trainer used to train an acoustic model from transcriptions using the EM algorithm. This data can only be used to train the symbol sequences stored in seqs during the creation of the auxiliary data. Currently supported types are: Bernoulli, DGaussian and Mixture.

Parameters:
amodelThe acoustic model.
countsThe logarithmic counts which will be increased.
seqsA symbol sequence manager containing the symbol sequences which can be used to train samples.
occsIf not NULL, it must be a pointer to a matrix in which the state occupancy values will be accumulated.
errPointer to string variable. If not NULL, an error message is allocated in the variable in case of error.
Returns:
Trainer, or NULL in case of error.
 All Data Structures Variables