libTLK  1.3.1
Functions
CMLLR adaptation
Algorithms

Functions

 tL_cmllr_adapt_fea (tLCMLLR *cmllr, const tLFea *fea_in, tLFea *fea_out)
 Adapt a feature sequence.
 tL_cmllr_free (tLCMLLR *cmllr)
 Frees memory.
 tL_cmllr_get_lv_constant (const tLCMLLR *cmllr, const tLAMlogCounts *counts)
 Gets the additional $\frac{1}{2}\sum_m\sum_t\log{(|A|)}$ term.
 tL_cmllr_init (tLCMLLR *cmllr)
 Initializes the CMLLR transformation matrix.
 tL_cmllr_new (tLAModel *amodel, char **err)
 Get a tLCMLLR.
 tL_cmllr_print_matrix (const tLCMLLR *cmllr, FILE *to)
 Prints the transformation matrix.
 tL_cmllr_read_matrix (const tLCMLLR *cmllr, gzFile from, char **err)
 Reads the transformation matrix.
 tL_cmllr_update_matrix (tLCMLLR *cmllr, const tLAMlogCounts *counts, const unsigned int niters)
 Update the transformation matrix.

Function Documentation

tL_cmllr_adapt_fea ( tLCMLLR *  cmllr,
const tLFea fea_in,
tLFea fea_out 
)

Adapt a feature sequence.

This function applies the CMLLR transformation matrix to the feature sequence stored in the provided feature sequence manager. The feature sequence is assumed to be real and to have the same dimension as the acoustic model related to the CMLLR structure.

Parameters:
cmllrThe CMLLR structure.
fea_inA feature sequence manager containing the real feature sequence to adapt.
fea_outThe feature sequence manager to which the transformed sequence will be stored.
tL_cmllr_free ( tLCMLLR *  cmllr)

Frees memory.

Frees the memory allocated for the CMLLR structure.

Parameters:
cmllrThe CMLLR structure.
tL_cmllr_get_lv_constant ( const tLCMLLR *  cmllr,
const tLAMlogCounts counts 
)

Gets the additional $\frac{1}{2}\sum_m\sum_t\log{(|A|)}$ term.

This function calculates the additional term $\frac{1}{2}\sum_m\sum_t\log{(|A|)}$, which must be added to log-likelihood.

Parameters:
cmllrThe CMLLR structure.
countsLogarithmic counts related to the acoustic model of the CMLLR structure.
Returns:
The additional term.
tL_cmllr_init ( tLCMLLR *  cmllr)

Initializes the CMLLR transformation matrix.

This function initializes the CMLLR transformation matrix. The A matrix is initialized as the identity matrix, and the b vector is initialized with zeroes.

Parameters:
cmllrThe CMLLR structure.
tL_cmllr_new ( tLAModel amodel,
char **  err 
)

Get a tLCMLLR.

This function creates a CMLLR structure used to adapt models and transform features using the CMLLR technique. Only DGaussian based types are supported. This function calls 'tL_cmllr_init'.

Parameters:
amodelThe acoustic model.
errPointer to string variable. If not NULL, an error message is allocated in the variable in case of error.
Returns:
CMLLR structure, or NULL in case of error.
tL_cmllr_print_matrix ( const tLCMLLR *  cmllr,
FILE *  to 
)

Prints the transformation matrix.

This function prints the CMLLR transformation matrix to the specified file.

Parameters:
cmllrThe CMLLR structure.
toFile to which the matrix will be written.
tL_cmllr_read_matrix ( const tLCMLLR *  cmllr,
gzFile  from,
char **  err 
)

Reads the transformation matrix.

This function reads the CMLLR transformation matrix from the specified file.

Parameters:
cmllrThe CMLLR structure.
fromFile where the matrix is stored.
errPointer to string variable. If not NULL, an error message is allocated in the variable in case of error.
Returns:
-1 if error, 0 if success.
tL_cmllr_update_matrix ( tLCMLLR *  cmllr,
const tLAMlogCounts counts,
const unsigned int  niters 
)

Update the transformation matrix.

This function recalculates the CMLLR transformation matrix, using the previous values and the provided logarithmic counts. When niters is 0, diagonal CMLLR matrix is calculated. When full matrix is calculated, counts must be created using 'tL_cmlogcounts_new_cmllr'.

Parameters:
cmllrThe CMLLR structure.
countsLogarithmic counts related to the acoustic model of the CMLLR structure. The counts must be created using 'tL_cmlogcounts_new_cmllr' if niters>0.
nitersThe number of iterations for the CMLLR, or 0 to indicate that diagonal CMLLR matrix must be calculated. algorithm. Must be greater than 0.
 All Data Structures Variables