ref: de6e99e8fd5210ec77df16b53da7618b61aa7f9d
dir: /include/cst_cg.h/
/*************************************************************************/
/* */
/* Language Technologies Institute */
/* Carnegie Mellon University */
/* Copyright (c) 2007-2017 */
/* All Rights Reserved. */
/* */
/* Permission is hereby granted, free of charge, to use and distribute */
/* this software and its documentation without restriction, including */
/* without limitation the rights to use, copy, modify, merge, publish, */
/* distribute, sublicense, and/or sell copies of this work, and to */
/* permit persons to whom this work is furnished to do so, subject to */
/* the following conditions: */
/* 1. The code must retain the above copyright notice, this list of */
/* conditions and the following disclaimer. */
/* 2. Any modifications must be clearly marked as such. */
/* 3. Original authors' names are not deleted. */
/* 4. The authors' names are not used to endorse or promote products */
/* derived from this software without specific prior written */
/* permission. */
/* */
/* CARNEGIE MELLON UNIVERSITY AND THE CONTRIBUTORS TO THIS WORK */
/* DISCLAIM ALL WARRANTIES WITH REGARD TO THIS SOFTWARE, INCLUDING */
/* ALL IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS, IN NO EVENT */
/* SHALL CARNEGIE MELLON UNIVERSITY NOR THE CONTRIBUTORS BE LIABLE */
/* FOR ANY SPECIAL, INDIRECT OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES */
/* WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN */
/* AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, */
/* ARISING OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE OF */
/* THIS SOFTWARE. */
/* */
/*************************************************************************/
/* Author: Alan W Black (awb@cs.cmu.edu) */
/* Date: November 2007 */
/*************************************************************************/
/* */
/* clustergen db */
/* */
/* A statistical corpus based synthesizer. */
/* See Black, A. (2006), CLUSTERGEN: A Statistical Parametric */
/* Synthesizer using Trajectory Modeling", Interspeech 2006 - ICSLP, */
/* Pittsburgh, PA. */
/* http://www.cs.cmu.edu/~awb/papers/is2006/IS061394.PDF */
/* */
/*************************************************************************/
#ifndef _CST_CG_H__
#define _CST_CG_H__
#include "cst_cart.h"
#include "cst_track.h"
#include "cst_wave.h"
#include "cst_audio.h"
#include "cst_synth.h" /* for dur_stat */
#include "cst_phoneset.h"
/* The parameters in the spectral model(s) may be encoded in various ways */
/* mostly to preserve space. The decoder can be dependent on the shape type */
/* This needs to be coordinated with the model creation in make_cg.scm */
/* This is currently very BIG_HAMMER-like and hard coded */
/* Original rf3 (and before) encoding stddev follows mean in pairs:
F0 (1), MCEP_STATIC (25), MCEP_DELTA (25), ME (5), Voicing (1) (114 shorts)
Encoded as linear range in unsigned short from min over range */
#define CST_CG_MODEL_SHAPE_BASE_MINRANGE 1
/* Quantized rf3, encode everything as two 8 bit values in each short.
Good generic quantized version, but the next one is used for most models
(57 shorts)
*/
#define CST_CG_MODEL_SHAPE_QUANTIZED_PARAMS 2
/* Quantized rf3, (mostly stddev follows mean in pairs -- except deltas
no F0, MCEP_STATIC (50) MCEP_DELTAS stddev (25), ME (5), Voice (2)
Encoded as unsigned chars that index into qtable. we don't include
the mcep_deltas means, they aren't needed. They are still held as shorts
so you need to sub index them see cst_cg_quantized_params_index()
(41 shorts) */
#define CST_CG_MODEL_SHAPE_QUANTIZED_PARAMS_41 3
typedef struct cst_cg_db_struct {
/* Please do not change this structure, but if you do, only add things
to the end of the struct. If do you change it , please modify
dump/load voice too (in cst_cg_dump_voice and cst_cg_map) */
const char *name;
const char * const *types;
int num_types;
int sample_rate;
float f0_mean, f0_stddev;
/* Cluster trees */
int num_f0_models;
const cst_cart ***f0_trees;
int num_param_models;
const cst_cart *** param_trees;
const cst_cart *spamf0_accent_tree; /* spam accent tree */
const cst_cart *spamf0_phrase_tree; /* spam phrase tree */
/* Model params e.g. mceps, deltas intersliced with stddevs */
/* may be compressed/quantized based on value of model_shape */
int *num_channels;
int *num_frames;
const unsigned short *** model_vectors;
int num_channels_spamf0_accent;
int num_frames_spamf0_accent;
const float * const * spamf0_accent_vectors;
/* Currently shared between different models */
const float *model_min; /* for vector coeffs encoding */
const float *model_range; /* for vector coeffs encoding */
const float ***qtable; /* q(uantization) tables for each model */
int model_shape; /* model compression technique */
float frame_advance;
/* duration models (cart + phonedurs) */
int num_dur_models;
const dur_stat *** dur_stats;
const cst_cart ** dur_cart;
/* phone to states map */
const char * const * const *phone_states;
/* Other parameters */
int do_mlpg; /* implies deltas are in the model_vectors */
float *dynwin;
int dynwinsize;
float mlsa_alpha;
float mlsa_beta;
int multimodel;
int mixed_excitation;
/* filters for Mixed Excitation */
int ME_num;
int ME_order;
const double * const *me_h;
int spamf0;
float gain;
/* If its a "grapheme" voice there will be a phoneset and a char_map */
const cst_phoneset *phoneset;
const char * const *char_sym_map; /* Unicode char to symbol map */
const char * const *sym_phone_map; /* symbol to phone map */
int freeable; /* doesn't get dumped, but 1 when this a freeable struct */
} cst_cg_db;
CST_VAL_USER_TYPE_DCLS(cg_db,cst_cg_db)
void delete_cg_db(cst_cg_db *db);
cst_utterance *cg_synth(cst_utterance *utt);
cst_wave *mlsa_resynthesis(const cst_track *t,
const cst_track *str,
cst_cg_db *cg_db,
cst_audio_streaming_info *asc,
int mlsa_speech_param);
cst_track *mlpg(const cst_track *param_track, cst_cg_db *cg_db);
cst_voice *cst_cg_load_voice(const char *voxdir,
const cst_lang lang_table[]);
int cst_cg_dump_voice(const cst_voice *v,const cst_string *filename);
#endif