subvq.h File Reference

Implementation of Sub-vector quantization. More...

#include <cmd_ln.h>
#include <logmath.h>
#include "s3types.h"
#include "cont_mgau.h"
#include "vector.h"

Go to the source code of this file.

Classes
struct	subvq_t
	Wrapper structures of sub-vector quantization. More...
Functions
subvq_t *	subvq_init (const char *file, float64 varfloor, int32 max_sv, mgau_model_t *g, cmd_ln_t *config, logmath_t *logmath)
void	subvq_free (subvq_t *vq)
int32	subvq_frame_eval (subvq_t *vq, mgau_model_t *g, int32 beam, float32 *feat, int32 *sen_active, int32 *senscr)
void	subvq_gautbl_eval_logs3 (subvq_t *vq, float32 *feat, logmath_t *logmath)
void	subvq_subvec_eval_logs3 (subvq_t *vq, float32 *feat, int32 sv, logmath_t *logmath)
int32	subvq_mgau_shortlist (subvq_t *vq, int32 m, int32 n, int32 beam)
int32	subvq_mgau_eval (mgau_model_t *g, subvq_t *vq, int32 m, int32 n, int32 *active)

---

Detailed Description

Implementation of Sub-vector quantization.

---

Function Documentation

int32 subvq_frame_eval (  subvq_t *  vq, mgau_model_t *  g, int32  beam, float32 *  feat, int32 *  sen_active, int32 *  senscr )

Evaluate senone scores for one frame. If subvq model is available, for each senone, first get approximate Gaussian density scores using it; obtain a shortlist of Gaussians using these scores, then evaluate the shortlist exactly. If no subvq model, evaluate senones using all Gaussian densities. Finally, scale senone scores by subtracting the best. Returns: The normalization factor (best senone absolute score). Parameters: vq  In: A sub-vector model g  In: Exact mixture Gaussian model beam  In: (Logs3) threshold for selecting shortlist; range = [-infinity(widest beam), 0(narrowest)] feat  In: Input feature vector for this frame sen_active  In: Active flags for each senone (optional). If not NULL, only active ones evaluated senscr  Out: Normalized senone scores

void subvq_free (  subvq_t *  vq  )

Deallocate sub-vector quantization Parameters: vq  In: A sub-vector model

void subvq_gautbl_eval_logs3 (  subvq_t *  vq, float32 *  feat, logmath_t *  logmath )

Evaluate the Mahalanobis distances between the given feature vector and each entry in the given subvq codebook. Save results, as logs3 values, in vq->vqdist[][]. Parameters: vq  In/Out: Reference subvq structure feat  In: Subvectors extracted from this, and compared to relevant subvq codewords

subvq_t* subvq_init (  const char *  file, float64  varfloor, int32  max_sv, mgau_model_t *  g, cmd_ln_t *  config, logmath_t *  logmath )

SubVQ file format: VQParam #Original-Codebooks #Original-Codewords/codebook(max) -> #Subvectors #VQ-codewords Subvector 0 length <length> <feature-dim> <feature-dim> <feature-dim> ... Subvector 1 length <length> <feature-dim> <feature-dim> <feature-dim> ... ... Codebook 0 Row 0 of mean/var values (interleaved) for subvector 0 codebook (in 1 line) Row 1 of above Row 2 of above ... Map 0 Mappings for state 0 codewords (in original model) to codewords of this subvector codebook Mappings for state 1 codewords (in original model) to codewords of this subvector codebook Mappings for state 2 codewords (in original model) to codewords of this subvector codebook ... Repeated for each subvector codebook 1 Repeated for each subvector codebook 2 ... End Returns: initialized sub-vq Parameters: file  In: Subvector model file varfloor  In: Floor to be applied to variance values max_sv  In: Use the first so many subvectors instead of all; if <0, use all g  In: Original model from which this subvq model was built, for cross-validation; optional

int32 subvq_mgau_eval (  mgau_model_t *  g, subvq_t *  vq, int32  m, int32  n, int32 *  active )

Compute the scores of a gaussian using only sum of the sub-vector scores. Returns: best senone score. Parameters: g  In: Gaussian vq  In: the SVQ m  In: GMM Index n  In :#Components in a specified mixture active  In:Active list of mixture

int32 subvq_mgau_shortlist (  subvq_t *  vq, int32  m, int32  n, int32  beam )

Parameters: vq  In subvq m  In: GMM index n  In: #Components in specified mixture beam  In: Threshold to select active components

void subvq_subvec_eval_logs3 (  subvq_t *  vq, float32 *  feat, int32  sv, logmath_t *  logmath )

Evaluate the codewords for a single given subvector sv, wrt the input feature vector. Save results, as logs3 values, in vq->vqdist[sv][]. (Basically, like subvq_gautbl_eval_logs3, but for a single given subvector instead of all.) Parameters: vq  In/Out: Reference subvq structure feat  In: Input feature subvector extracted from this, and compared to relevant codewords sv  In: ID of subvector being evaluated

---