ms_mllr.h File Reference

(Sphinx 3.0 specific) Multiple stream MLLR. More...

#include <s3types.h>

Go to the source code of this file.

Functions
int32	ms_mllr_read_regmat (const char *regmatfile, float32 *****A, float32 ****B, float32 ****H, int32 *streamlen, int32 n_stream, int32 *nclass)
int32	ms_mllr_free_regmat (float32 ****A, float32 ***B, float32 ***H, int32 n_stream)
int32	ms_mllr_norm_mgau (float32 ***mean, float32 ***mvar, int32 n_density, float32 ****A, float32 ***B, float32 ***H, int32 *streamlen, int32 n_stream, int32 _class)

---

Detailed Description

(Sphinx 3.0 specific) Multiple stream MLLR.

---

Function Documentation

int32 ms_mllr_free_regmat (  float32 ****  A, float32 ***  B, float32 ***  H, int32  n_stream )

Free a regression matrix previously read in by mllr_read_regmat. Return value: 0 if successful, -1 otherwise. Parameters: A  In: A[stream][nclass][streamlen][streamlen] B  In: B[stream][nclass][streamlen] H  In: H[stream][nclass][streamlen] n_stream  In: #Feature streams

int32 ms_mllr_norm_mgau (  float32 ***  mean, float32 ***  mvar, int32  n_density, float32 ****  A, float32 ***  B, float32 ***  H, int32 *  streamlen, int32  n_stream, int32  _class )

Transform a mixture density mean matrix according to the given regression matrix. Return value: 0 if successful, -1 otherwise. Parameters: mean  < In/Out: Set of vectors to be transformed: mean[stream][density] mvar  < In/Out: Set of vectors to be transformed: var[stream][density] n_density  In: #densities in above mean matrix A  In: "matrix" portion of regression matrix B  In: "vector" portion of regression matrix H  In: regression "matrix" for variances streamlen  In: Dimension of each feature stream n_stream  In: #Feature streams _class  In: number of classes

int32 ms_mllr_read_regmat (  const char *  regmatfile, float32 *****  A, float32 ****  B, float32 ****  H, int32 *  streamlen, int32  n_stream, int32 *  nclass )

Load a regression matrix from the given file. Space for the matrix is allocated by this routine. (The regression "matrix" is actually a matrix A and a vector B.) Return value: 0 if successful, -1 otherwise. Parameters: regmatfile  In: File to be read A  Out: [*A][stream][nclass][streamlen][streamlen] B  Out: [*B][stream][nclass][streamlen] H  Out: [*H][stream][nclass][streamlen] streamlen  In: Dimension of each feature stream n_stream  In: #Feature streams nclass  Out: number of classes

---