/usr/include/sphinx3/subvq.h is in libs3decoder-dev 0.8-0ubuntu1.
This file is owned by root:root, with mode 0o644.
The actual contents of the file can be viewed below.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 | /* -*- c-basic-offset: 4; indent-tabs-mode: nil -*- */
/* ====================================================================
* Copyright (c) 1999-2004 Carnegie Mellon University. All rights
* reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
*
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
*
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in
* the documentation and/or other materials provided with the
* distribution.
*
* This work was supported in part by funding from the Defense Advanced
* Research Projects Agency and the National Science Foundation of the
* United States of America, and the CMU Sphinx Speech Consortium.
*
* THIS SOFTWARE IS PROVIDED BY CARNEGIE MELLON UNIVERSITY ``AS IS'' AND
* ANY EXPRESSED OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
* THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
* PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL CARNEGIE MELLON UNIVERSITY
* NOR ITS EMPLOYEES BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
* SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
* LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
* DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
* THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*
* ====================================================================
*
*/
/*
* subvq.h
*
* **********************************************
* CMU ARPA Speech Project
*
* Copyright (c) 1999 Carnegie Mellon University.
* ALL RIGHTS RESERVED.
* **********************************************
*
* HISTORY
* $Log$
* Revision 1.1 2006/04/05 20:27:30 dhdfu
* A Great Reorganzation of header files and executables
*
* Revision 1.14 2006/02/22 17:43:32 arthchan2003
* Merged from SPHINX3_5_2_RCI_IRII_BRANCH:
* 1, vector_gautbl_free is not appropiate to be used in this case because it will free a certain piece of memory twice.
* 2, Fixed dox-doc.
*
* Revision 1.13.4.1 2005/07/05 05:47:59 arthchan2003
* Fixed dox-doc. struct level of documentation are included.
*
* Revision 1.13 2005/06/21 19:01:33 arthchan2003
* Added $ keyword.
*
* Revision 1.3 2005/03/30 01:22:47 archan
* Fixed mistakes in last updates. Add
*
*
* 20.Apr.2001 RAH (rhoughton@mediasite.com, ricky.houghton@cs.cmu.edu)
* Updated subvq_free () to free more allocated memory
*
* 15-Dec-1999 M K Ravishankar (rkm@cs.cmu.edu) at Carnegie Mellon University
* Moved subvq_t.{frm_sen_eval,frm_gau_eval} to cont_mgau.h.
*
* 14-Dec-1999 M K Ravishankar (rkm@cs.cmu.edu) at Carnegie Mellon University
* Added subvq_t.{frm_sen_eval,frm_gau_eval}. Changed subvq_frame_eval to
* return the normalization factor.
*
* 06-Dec-1999 M K Ravishankar (rkm@cs.cmu.edu) at Carnegie Mellon University
* Added subvq_subvec_eval_logs3().
*
* 14-Oct-1999 M K Ravishankar (rkm@cs.cmu.edu) at Carnegie Mellon University
* Changed ci_active flags input to sen_active in subvq_frame_eval().
*
* 20-Jul-1999 M K Ravishankar (rkm@cs.cmu.edu) at Carnegie Mellon University
* Added subvq_gautbl_eval_logs3().
*
* 12-Mar-1999 M K Ravishankar (rkm@cs.cmu.edu) at Carnegie Mellon University
* Started.
*/
#ifndef _S3_SUBVQ_H_
#define _S3_SUBVQ_H_
#include <cmd_ln.h>
#include <logmath.h>
#include "s3types.h"
#include "cont_mgau.h"
#include "vector.h"
/** \file subvq.h
\brief Implementation of Sub-vector quantization.
*/
#ifdef __cplusplus
extern "C" {
#endif
#if 0
} /* Fool Emacs into not indenting things. */
#endif
/** \struct subvq_t
* \brief Wrapper structures of sub-vector quantization
*/
typedef struct {
arraysize_t origsize; /**< origsize.r = #codebooks (or states) in original model;
origsize.c = max #codewords/codebook in original model. */
int32 n_sv; /**< #Subvectors */
int32 vqsize; /**< #Codewords in each subvector quantized mean/var table */
int32 **featdim; /**< featdim[s] = Original feature dimensions in subvector s */
vector_gautbl_t *gautbl; /**< Vector-quantized Gaussians table for each sub-vector */
int32 ***map; /**< map[i][j] = map from original codebook(i)/codeword(j) to
sequence of nearest vector quantized subvector codewords;
so, each map[i][j] is of length n_sv. Finally, map is
LINEARIZED, so that it indexes into a 1-D array of scores
rather than a 2-D array (for faster access). */
/* Working space used during evaluation. */
float32 *subvec; /**< Subvector extracted from feature vector */
int32 **vqdist; /**< vqdist[i][j] = score (distance) for i-th subvector compared
to j-th subvector-codeword */
int32 *gauscore; /**< Subvq-based approx. Gaussian density scores for one mixture */
int32 *mgau_sl; /**< Shortlist for one mixture (based on gauscore[]) */
/* ARCHAN, 1111, 04, move the static global variables to the structure again. */
/* RAH, 5.8.01, VQ_EVAL determines how many vectors are used to
* compute the shortlist, for now this value is only relevant when n_sv =3.
* Setting it to 1 means that only the CEP values are estimated, 2 means that
* CEP and delta values are estimated, 3 means all three are estimated.
* Note, we must adjust the beam widths as we muck around with these.
*/
int32 VQ_EVAL; /** Number of sub-vector to be computed */
} subvq_t;
/**
* 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
* @return initialized sub-vq
*/
subvq_t *subvq_init (const char *file, /**< In: Subvector model file */
float64 varfloor, /**< In: Floor to be applied to variance values */
int32 max_sv, /**< In: Use the first so many subvectors instead of all;
if <0, use all */
mgau_model_t *g, /**< In: Original model from
which this subvq model was
built, for
cross-validation; optional */
cmd_ln_t *config,
logmath_t *logmath
);
/** Deallocate sub-vector quantization */
void subvq_free (subvq_t *vq /**< In: A sub-vector model */
);
/**
* 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.
* @return The normalization factor (best senone absolute score).
*/
int32 subvq_frame_eval (subvq_t *vq, /**< In: A sub-vector model */
mgau_model_t *g,/**< In: Exact mixture Gaussian model */
int32 beam, /**< In: (Logs3) threshold for selecting shortlist;
range = [-infinity(widest beam), 0(narrowest)] */
float32 *feat, /**< In: Input feature vector for this frame */
int32 *sen_active, /**< In: Active flags for each senone (optional).
If not NULL, only active ones evaluated */
int32 *senscr /**< Out: Normalized senone scores */
);
/**
* 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[][].
*/
void subvq_gautbl_eval_logs3 (subvq_t *vq, /**< In/Out: Reference subvq structure */
float32 *feat, /**< In: Subvectors
extracted from this, and compared to
relevant subvq codewords */
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.)
*/
void subvq_subvec_eval_logs3 (subvq_t *vq, /**< In/Out: Reference subvq structure */
float32 *feat, /**< In: Input feature
** subvector
** extracted from
** this, and compared
** to relevant
** codewords */
int32 sv, /**< In: ID of
subvector being
evaluated */
logmath_t *logmath
);
/*
* Based on previously computed subvq scores (Mahalanobis distances), determine the active
* components in the given mixture (using the vq->map).
* @return Number of candidates in the returned shortlist.
*/
int32 subvq_mgau_shortlist (subvq_t *vq, /**< In subvq */
int32 m, /**< In: GMM index */
int32 n, /**< In: #Components in specified mixture */
int32 beam /**< In: Threshold to select active components */
);
/**
* Compute the scores of a gaussian using only sum of the sub-vector scores.
* @return best senone score.
*/
int32 subvq_mgau_eval (mgau_model_t *g, /**< In: Gaussian */
subvq_t *vq, /**< In: the SVQ */
int32 m, /**< In: GMM Index */
int32 n, /**< In :#Components in a specified mixture */
int32 *active /**< In:Active list of mixture */
);
#if 0
{ /* Stop indent from complaining */
#endif
#ifdef __cplusplus
}
#endif
#endif
|