HomoFast-eSpeak-Persian/src/libespeak-ng/voices.c

/*
 * Copyright (C) 2005 to 2015 by Jonathan Duddington
 * email: [email protected]
 * Copyright (C) 2015-2017 Reece H. Dunn
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 3 of the License, or
 * (at your option) any later version.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, see: <http://www.gnu.org/licenses/>.
 */

#include "config.h"

#include <ctype.h>
#include <wctype.h>
#include <errno.h>
#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <strings.h>

#if defined(_WIN32) || defined(_WIN64)
#include <windows.h>
#else
#include <dirent.h>
#endif

#include <espeak-ng/espeak_ng.h>
#include <espeak-ng/speak_lib.h>
#include <espeak-ng/encoding.h>

#include "speech.h"
#include "phoneme.h"
#include "voice.h"
#include "synthesize.h"
#include "translate.h"

MNEM_TAB genders[] = {
	{ "male", ENGENDER_MALE },
	{ "female", ENGENDER_FEMALE },
	{ NULL, ENGENDER_MALE }
};

int tone_points[12] = { 600, 170, 1200, 135, 2000, 110, 3000, 110, -1, 0 };

// limit the rate of change for each formant number
static int formant_rate_22050[9] = { 240, 170, 170, 170, 170, 170, 170, 170, 170 }; // values for 22kHz sample rate
int formant_rate[9]; // values adjusted for actual sample rate

#define DEFAULT_LANGUAGE_PRIORITY  5
#define N_VOICES_LIST  250
static int n_voices_list = 0;
static espeak_VOICE *voices_list[N_VOICES_LIST];

espeak_VOICE current_voice_selected;

enum {
	V_NAME = 1,
	V_LANGUAGE,
	V_GENDER,
	V_TRANSLATOR,
	V_PHONEMES,
	V_DICTIONARY,
	V_VARIANTS,

	V_MAINTAINER,
	V_STATUS,

	// these affect voice quality, are independent of language
	V_FORMANT,
	V_PITCH,
	V_ECHO,
	V_FLUTTER,
	V_ROUGHNESS,
	V_CLARITY,
	V_TONE,
	V_VOICING,
	V_BREATH,
	V_BREATHW,

	// these override defaults set by the translator
	V_WORDGAP,
	V_INTONATION,
	V_TUNES,
	V_STRESSLENGTH,
	V_STRESSAMP,
	V_STRESSADD,
	V_DICTRULES,
	V_STRESSRULE,
	V_STRESSOPT,
	V_NUMBERS,
	V_OPTION,

	V_MBROLA,
	V_KLATT,
	V_FAST,
	V_SPEED,
	V_DICTMIN,
	V_ALPHABET2,

	// these need a phoneme table to have been specified
	V_REPLACE,
	V_CONSONANTS,

	// these are alpha features that need to be tested and categorized
	V_LETTER_VOWEL

};

static MNEM_TAB options_tab[] = {
	{ "reduce_t", LOPT_REDUCE_T },
	{ "bracket",  LOPT_BRACKET_PAUSE },
	{ NULL,   -1 }
};

static MNEM_TAB keyword_tab[] = {
	{ "name",         V_NAME },
	{ "language",     V_LANGUAGE },
	{ "gender",       V_GENDER },

	{ "maintainer",   V_MAINTAINER },
	{ "status",       V_STATUS },

	{ "variants",     V_VARIANTS },
	{ "formant",      V_FORMANT },
	{ "pitch",        V_PITCH },
	{ "phonemes",     V_PHONEMES },
	{ "translator",   V_TRANSLATOR },
	{ "dictionary",   V_DICTIONARY },
	{ "stressLength", V_STRESSLENGTH },
	{ "stressAmp",    V_STRESSAMP },
	{ "stressAdd",    V_STRESSADD },
	{ "intonation",   V_INTONATION },
	{ "tunes",        V_TUNES },
	{ "dictrules",    V_DICTRULES },
	{ "stressRule",   V_STRESSRULE },
	{ "stressopt",    V_STRESSOPT },
	{ "replace",      V_REPLACE },
	{ "words",        V_WORDGAP },
	{ "echo",         V_ECHO },
	{ "flutter",      V_FLUTTER },
	{ "roughness",    V_ROUGHNESS },
	{ "clarity",      V_CLARITY },
	{ "tone",         V_TONE },
	{ "voicing",      V_VOICING },
	{ "breath",       V_BREATH },
	{ "breathw",      V_BREATHW },
	{ "numbers",      V_NUMBERS },
	{ "option",       V_OPTION },
	{ "mbrola",       V_MBROLA },
	{ "consonants",   V_CONSONANTS },
	{ "klatt",        V_KLATT },
	{ "fast_test2",   V_FAST },
	{ "speed",        V_SPEED },
	{ "dict_min",     V_DICTMIN },

	// these just set a value in langopts.param[]
	{ "l_dieresis",       0x100+LOPT_DIERESES },
	{ "l_prefix",         0x100+LOPT_PREFIXES },
	{ "l_regressive_v",   0x100+LOPT_REGRESSIVE_VOICING },
	{ "l_unpronouncable", 0x100+LOPT_UNPRONOUNCABLE },
	{ "l_sonorant_min",   0x100+LOPT_SONORANT_MIN },
	{ "l_length_mods",    0x100+LOPT_LENGTH_MODS },
	{ "apostrophe",       0x100+LOPT_APOSTROPHE },

	// these are alpha features that need to be tested and categorized
	{ "letterVowel", V_LETTER_VOWEL },
	{ NULL, 0 }
};

#define N_VOICE_VARIANTS   12
const char variants_either[N_VOICE_VARIANTS] = { 1, 2, 12, 3, 13, 4, 14, 5, 11, 0 };
const char variants_male[N_VOICE_VARIANTS] = { 1, 2, 3, 4, 5, 6, 0 };
const char variants_female[N_VOICE_VARIANTS] = { 11, 12, 13, 14, 0 };
const char *variant_lists[3] = { variants_either, variants_male, variants_female };

static voice_t voicedata;
voice_t *voice = &voicedata;

static char *fgets_strip(char *buf, int size, FILE *f_in)
{
	// strip trailing spaces, and truncate lines at // comment
	int len;
	char *p;

	if (fgets(buf, size, f_in) == NULL)
		return NULL;

	if (buf[0] == '#') {
		buf[0] = 0;
		return buf;
	}

	len = strlen(buf);
	while ((--len > 0) && isspace(buf[len]))
		buf[len] = 0;

	if ((p = strstr(buf, "//")) != NULL)
		*p = 0;

	return buf;
}

static int LookupTune(const char *name)
{
	int ix;

	for (ix = 0; ix < n_tunes; ix++) {
		if (strcmp(name, tunes[ix].name) == 0)
			return ix;
	}
	return -1;
}

static void SetToneAdjust(voice_t *voice, int *tone_pts)
{
	int ix;
	int pt;
	int y;
	int freq1 = 0;
	int freq2;
	int height1 = tone_pts[1];
	int height2;
	double rate;

	for (pt = 0; pt < 12; pt += 2) {
		if (tone_pts[pt] == -1) {
			tone_pts[pt] = N_TONE_ADJUST*8;
			if (pt > 0)
				tone_pts[pt+1] = tone_pts[pt-1];
		}
		freq2 = tone_pts[pt] / 8; // 8Hz steps
		height2 = tone_pts[pt+1];
		if ((freq2 - freq1) > 0) {
			rate = (double)(height2-height1)/(freq2-freq1);

			for (ix = freq1; ix < freq2; ix++) {
				y = height1 + (int)(rate * (ix-freq1));
				if (y > 255)
					y = 255;
				voice->tone_adjust[ix] = y;
			}
		}
		freq1 = freq2;
		height1 = height2;
	}
}

void ReadTonePoints(char *string, int *tone_pts)
{
	// tone_pts[] is int[12]
	int ix;

	for (ix = 0; ix < 12; ix++)
		tone_pts[ix] = -1;

	sscanf(string, "%d %d %d %d %d %d %d %d %d %d",
	       &tone_pts[0], &tone_pts[1], &tone_pts[2], &tone_pts[3],
	       &tone_pts[4], &tone_pts[5], &tone_pts[6], &tone_pts[7],
	       &tone_pts[8], &tone_pts[9]);
}

static espeak_VOICE *ReadVoiceFile(FILE *f_in, const char *fname, int is_language_file)
{
	// Read a Voice file, allocate a VOICE_DATA and set data from the
	// file's  language, gender, name  lines

	char linebuf[120];
	char vname[80];
	char vgender[80];
	char vlanguage[80];
	char languages[300]; // allow space for several alternate language names and priorities

	unsigned int len;
	int langix = 0;
	int n_languages = 0;
	char *p;
	espeak_VOICE *voice_data;
	int priority;
	int age;
	int n_variants = 4; // default, number of variants of this voice before using another voice
	int gender;

	vname[0] = 0;
	vgender[0] = 0;
	age = 0;

	while (fgets_strip(linebuf, sizeof(linebuf), f_in) != NULL) {
		// isolate the attribute name
		for (p = linebuf; (*p != 0) && !iswspace(*p); p++) ;
		*p++ = 0;

		if (linebuf[0] == 0) continue;

		switch (LookupMnem(keyword_tab, linebuf))
		{
		case V_NAME:
			while (isspace(*p)) p++;
			strncpy0(vname, p, sizeof(vname));
			break;
		case V_LANGUAGE:
			priority = DEFAULT_LANGUAGE_PRIORITY;
			vlanguage[0] = 0;

			sscanf(p, "%s %d", vlanguage, &priority);
			len = strlen(vlanguage) + 2;
			// check for space in languages[]
			if (len < (sizeof(languages)-langix-1)) {
				languages[langix] = priority;

				strcpy(&languages[langix+1], vlanguage);
				langix += len;
				n_languages++;
			}
			break;
		case V_GENDER:
			sscanf(p, "%s %d", vgender, &age);
			if (is_language_file)
				fprintf(stderr, "Error (%s): gender attribute specified on a language file\n", fname);
			break;
		case V_VARIANTS:
			sscanf(p, "%d", &n_variants);
		}
	}
	languages[langix++] = 0;

	gender = LookupMnem(genders, vgender);

	if (n_languages == 0)
		return NULL; // no language lines in the voice file

	p = (char *)calloc(sizeof(espeak_VOICE) + langix + strlen(fname) + strlen(vname) + 3, 1);
	voice_data = (espeak_VOICE *)p;
	p = &p[sizeof(espeak_VOICE)];

	memcpy(p, languages, langix);
	voice_data->languages = p;

	strcpy(&p[langix], fname);
	voice_data->identifier = &p[langix];
	voice_data->name = &p[langix];

	if (vname[0] != 0) {
		langix += strlen(fname)+1;
		strcpy(&p[langix], vname);
		voice_data->name = &p[langix];
	}

	voice_data->age = age;
	voice_data->gender = gender;
	voice_data->variant = 0;
	voice_data->xx1 = n_variants;
	return voice_data;
}

void VoiceReset(int tone_only)
{
	// Set voice to the default values

	int pk;
	static unsigned char default_heights[N_PEAKS] = { 130, 128, 120, 116, 100, 100, 128, 128, 128 }; // changed for v.1.47
	static unsigned char default_widths[N_PEAKS] = { 140, 128, 128, 160, 171, 171, 128, 128, 128 };

	static int breath_widths[N_PEAKS] = { 0, 200, 200, 400, 400, 400, 600, 600, 600 };

	// default is:  pitch 80,118
	voice->pitch_base = 0x47000;
	voice->pitch_range = 4104;

	voice->formant_factor = 256;

	voice->speed_percent = 100;
	voice->echo_delay = 0;
	voice->echo_amp = 0;
	voice->flutter = 64;
	voice->n_harmonic_peaks = 5;
	voice->peak_shape = 0;
	voice->voicing = 64;
	voice->consonant_amp = 90; // change from 100 to 90 for v.1.47
	voice->consonant_ampv = 100;
	voice->samplerate = samplerate_native;
	memset(voice->klattv, 0, sizeof(voice->klattv));

	speed.fast_settings[0] = 450;
	speed.fast_settings[1] = 800;
	speed.fast_settings[2] = 175;

	voice->roughness = 2;

	InitBreath();
	for (pk = 0; pk < N_PEAKS; pk++) {
		voice->freq[pk] = 256;
		voice->height[pk] = default_heights[pk]*2;
		voice->width[pk] = default_widths[pk]*2;
		voice->breath[pk] = 0;
		voice->breathw[pk] = breath_widths[pk]; // default breath formant widths
		voice->freqadd[pk] = 0;

		// adjust formant smoothing depending on sample rate
		formant_rate[pk] = (formant_rate_22050[pk] * 22050)/samplerate;
	}

	// This table provides the opportunity for tone control.
	// Adjustment of harmonic amplitudes, steps of 8Hz
	// value of 128 means no change
	SetToneAdjust(voice, tone_points);

	// default values of speed factors
	voice->speedf1 = 256;
	voice->speedf2 = 238;
	voice->speedf3 = 232;

	if (tone_only == 0) {
		n_replace_phonemes = 0;
		LoadMbrolaTable(NULL, NULL, 0);
	}
}

static void VoiceFormant(char *p)
{
	// Set parameters for a formant
	int ix;
	int formant;
	int freq = 100;
	int height = 100;
	int width = 100;
	int freqadd = 0;

	ix = sscanf(p, "%d %d %d %d %d", &formant, &freq, &height, &width, &freqadd);
	if (ix < 2)
		return;

	if ((formant < 0) || (formant > 8))
		return;

	if (freq >= 0)
		voice->freq[formant] = (int)(freq * 2.56001);
	if (height >= 0)
		voice->height[formant] = (int)(height * 2.56001);
	if (width >= 0)
		voice->width[formant] = (int)(width * 2.56001);
	voice->freqadd[formant] = freqadd;
}

static void PhonemeReplacement(char *p)
{
	int n;
	int phon;
	int flags = 0;
	char phon_string1[12];
	char phon_string2[12];

	strcpy(phon_string2, "NULL");
	n = sscanf(p, "%d %s %s", &flags, phon_string1, phon_string2);
	if ((n < 2) || (n_replace_phonemes >= N_REPLACE_PHONEMES))
		return;

	if ((phon = LookupPhonemeString(phon_string1)) == 0)
		return; // not recognised

	replace_phonemes[n_replace_phonemes].old_ph = phon;
	replace_phonemes[n_replace_phonemes].new_ph = LookupPhonemeString(phon_string2);
	replace_phonemes[n_replace_phonemes++].type = flags;
}

static int Read8Numbers(char *data_in, int *data)
{
	// Read 8 integer numbers
	memset(data, 0, 8+sizeof(int));
	return sscanf(data_in, "%d %d %d %d %d %d %d %d",
	              &data[0], &data[1], &data[2], &data[3], &data[4], &data[5], &data[6], &data[7]);
}

voice_t *LoadVoice(const char *vname, int control)
{
	// control, bit 0  1= no_default
	//          bit 1  1 = change tone only, not language
	//          bit 2  1 = don't report error on LoadDictionary
	//          bit 4  1 = vname = full path

	FILE *f_voice = NULL;
	char *p;
	int key;
	int ix;
	int n;
	int value;
	int value2;
	int langix = 0;
	int tone_only = control & 2;
	bool language_set = false;
	bool phonemes_set = false;
	int stress_amps_set = 0;
	int stress_lengths_set = 0;
	int stress_add_set = 0;
	int conditional_rules = 0;
	LANGUAGE_OPTIONS *langopts = NULL;

	Translator *new_translator = NULL;

	char voicename[40];
	char language_name[40];
	char translator_name[40];
	char new_dictionary[40];
	char phonemes_name[40];
	char option_name[40];
	const char *language_type;
	char buf[sizeof(path_home)+30];
	char path_voices[sizeof(path_home)+12];

	int dict_min = 0;
	int stress_amps[8];
	int stress_lengths[8];
	int stress_add[8];
	char names[8][40];
	char name1[40];
	char name2[80];

	int pitch1;
	int pitch2;

	static char voice_identifier[40]; // file name for  current_voice_selected
	static char voice_name[40];       // voice name for current_voice_selected
	static char voice_languages[100]; // list of languages and priorities for current_voice_selected

	strncpy0(voicename, vname, sizeof(voicename));
	if (control & 0x10) {
		strcpy(buf, vname);
		if (GetFileLength(buf) <= 0)
			return NULL;
	} else {
		if (voicename[0] == 0)
			strcpy(voicename, ESPEAKNG_DEFAULT_VOICE);

		sprintf(path_voices, "%s%cvoices%c", path_home, PATHSEP, PATHSEP);
		sprintf(buf, "%s%s", path_voices, voicename); // look in the main voices directory

		if (GetFileLength(buf) <= 0) {
			sprintf(path_voices, "%s%clang%c", path_home, PATHSEP, PATHSEP);
			sprintf(buf, "%s%s", path_voices, voicename); // look in the main languages directory
		}
	}

	f_voice = fopen(buf, "r");

	language_type = "en"; // default
	if (f_voice == NULL) {
		if (control & 3)
			return NULL; // can't open file

		if (SelectPhonemeTableName(voicename) >= 0)
			language_type = voicename;
	}

	if (!tone_only && (translator != NULL)) {
		DeleteTranslator(translator);
		translator = NULL;
	}

	strcpy(translator_name, language_type);
	strcpy(new_dictionary, language_type);
	strcpy(phonemes_name, language_type);

	if (!tone_only) {
		voice = &voicedata;
		strncpy0(voice_identifier, vname, sizeof(voice_identifier));
		voice_name[0] = 0;
		voice_languages[0] = 0;

		current_voice_selected.identifier = voice_identifier;
		current_voice_selected.name = voice_name;
		current_voice_selected.languages = voice_languages;
	} else {
		// append the variant file name to the voice identifier
		if ((p = strchr(voice_identifier, '+')) != NULL)
			*p = 0;    // remove previous variant name
		sprintf(buf, "+%s", &vname[3]);    // omit  !v/  from the variant filename
		strcat(voice_identifier, buf);
		langopts = &translator->langopts;
	}
	VoiceReset(tone_only);

	if (!tone_only)
		SelectPhonemeTableName(phonemes_name); // set up phoneme_tab

	while ((f_voice != NULL) && (fgets_strip(buf, sizeof(buf), f_voice) != NULL)) {
		// isolate the attribute name
		for (p = buf; (*p != 0) && !isspace(*p); p++) ;
		*p++ = 0;

		if (buf[0] == 0) continue;

		key = LookupMnem(keyword_tab, buf);

		switch (key)
		{
		case V_LANGUAGE:
		{
			unsigned int len;
			int priority;

			if (tone_only)
				break;

			priority = DEFAULT_LANGUAGE_PRIORITY;
			language_name[0] = 0;

			sscanf(p, "%s %d", language_name, &priority);
			if (strcmp(language_name, "variant") == 0)
				break;

			len = strlen(language_name) + 2;
			// check for space in languages[]
			if (len < (sizeof(voice_languages)-langix-1)) {
				voice_languages[langix] = priority;

				strcpy(&voice_languages[langix+1], language_name);
				langix += len;
			}

			// only act on the first language line
			if (language_set == false) {
				language_type = strtok(language_name, "-");
				language_set = true;
				strcpy(translator_name, language_type);
				strcpy(new_dictionary, language_type);
				strcpy(phonemes_name, language_type);
				SelectPhonemeTableName(phonemes_name);

				if (new_translator != NULL)
					DeleteTranslator(new_translator);

				new_translator = SelectTranslator(translator_name);
				langopts = &new_translator->langopts;
				strncpy0(voice->language_name, language_name, sizeof(voice->language_name));
			}
		}
			break;
		case V_NAME:
			if (tone_only == 0) {
				while (isspace(*p)) p++;
				strncpy0(voice_name, p, sizeof(voice_name));
			}
			break;
		case V_GENDER:
		{
			int age = 0;
			char vgender[80];
			sscanf(p, "%s %d", vgender, &age);
			current_voice_selected.gender = LookupMnem(genders, vgender);
			current_voice_selected.age = age;
		}
			break;
		case V_TRANSLATOR:
			if (tone_only) break;

			sscanf(p, "%s", translator_name);

			if (new_translator != NULL)
				DeleteTranslator(new_translator);

			new_translator = SelectTranslator(translator_name);
			langopts = &new_translator->langopts;
			break;
		case V_DICTIONARY: // dictionary
			sscanf(p, "%s", new_dictionary);
			break;
		case V_PHONEMES: // phoneme table
			sscanf(p, "%s", phonemes_name);
			break;
		case V_FORMANT:
			VoiceFormant(p);
			break;
		case V_PITCH:
			// default is  pitch 82 118
			if (sscanf(p, "%d %d", &pitch1, &pitch2) == 2) {
				voice->pitch_base = (pitch1 - 9) << 12;
				voice->pitch_range = (pitch2 - pitch1) * 108;
				double factor = (double)(pitch1 - 82)/82;
				voice->formant_factor = (int)((1+factor/4) * 256); // nominal formant shift for a different voice pitch
			}
			break;
		case V_STRESSLENGTH: // stressLength
			stress_lengths_set = Read8Numbers(p, stress_lengths);
			break;
		case V_STRESSAMP: // stressAmp
			stress_amps_set = Read8Numbers(p, stress_amps);
			break;
		case V_STRESSADD: // stressAdd
			stress_add_set = Read8Numbers(p, stress_add);
			break;
		case V_INTONATION: // intonation
			sscanf(p, "%d %d", &option_tone_flags, &option_tone2);
			if ((option_tone_flags & 0xff) != 0) {
				if (langopts)
					langopts->intonation_group = option_tone_flags & 0xff;
				else
					fprintf(stderr, "Cannot set intonation: language not set, or is invalid.\n");
			}
			break;
		case V_TUNES:
			n = sscanf(p, "%s %s %s %s %s %s", names[0], names[1], names[2], names[3], names[4], names[5]);
			if (langopts) {
				langopts->intonation_group = 0;
				for (ix = 0; ix < n; ix++) {
					if (strcmp(names[ix], "NULL") == 0)
						continue;

					if ((value = LookupTune(names[ix])) < 0)
						fprintf(stderr, "Unknown tune '%s'\n", names[ix]);
					else
						langopts->tunes[ix] = value;
				}
			} else
				fprintf(stderr, "Cannot set tunes: language not set, or is invalid.\n");
			break;
		case V_DICTRULES: // conditional dictionary rules and list entries
		case V_NUMBERS:
		case V_STRESSOPT:
			if (langopts) {
				// expect a list of numbers
				while (*p != 0) {
					while (isspace(*p)) p++;
					if ((n = atoi(p)) > 0) {
						p++;
						if (n < 32) {
							if (key == V_DICTRULES)
								conditional_rules |= (1 << n);
							else if (key == V_NUMBERS)
								langopts->numbers |= (1 << n);
							else if (key == V_STRESSOPT)
								langopts->stress_flags |= (1 << n);
						} else {
							if ((key == V_NUMBERS) && (n < 64))
								langopts->numbers2 |= (1 << (n-32));
							else
								fprintf(stderr, "Bad option number %d\n", n);
						}
					}
					while (isalnum(*p)) p++;
				}
				ProcessLanguageOptions(langopts);
			} else
				fprintf(stderr, "Cannot set stressopt: language not set, or is invalid.\n");
			break;
		case V_REPLACE:
			if (phonemes_set == false) {
				// must set up a phoneme table before we can lookup phoneme mnemonics
				SelectPhonemeTableName(phonemes_name);
				phonemes_set = true;
			}
			PhonemeReplacement(p);
			break;
		case V_WORDGAP: // words
			if (langopts)
				sscanf(p, "%d %d", &langopts->word_gap, &langopts->vowel_pause);
			else
				fprintf(stderr, "Cannot set wordgap: language not set, or is invalid.\n");
			break;
		case V_STRESSRULE:
			if (langopts)
				sscanf(p, "%d %d %d %d", &langopts->stress_rule,
				       &langopts->stress_flags,
				       &langopts->unstressed_wd1,
				       &langopts->unstressed_wd2);
			else
				fprintf(stderr, "Cannot set stressrule: language not set, or is invalid.\n");
			break;
		case V_OPTION:
			if (langopts) {
				value2 = 0;
				if (((sscanf(p, "%s %d %d", option_name, &value, &value2) >= 2) && ((ix = LookupMnem(options_tab, option_name)) >= 0)) ||
				    ((sscanf(p, "%d %d %d", &ix, &value, &value2) >= 2) && (ix < N_LOPTS))) {
					langopts->param[ix] = value;
					langopts->param2[ix] = value2;
				} else
					fprintf(stderr, "Bad voice option: %s %s\n", buf, p);
			} else
				fprintf(stderr, "Cannot set option: language not set, or is invalid.\n");
			break;
		case V_ECHO:
			// echo.  suggest: 135mS  11%
			value = 0;
			voice->echo_amp = 0;
			sscanf(p, "%d %d", &voice->echo_delay, &voice->echo_amp);
			break;
		case V_FLUTTER: // flutter
			if (sscanf(p, "%d", &value) == 1)
				voice->flutter = value * 32;
			break;
		case V_ROUGHNESS: // roughness
			if (sscanf(p, "%d", &value) == 1)
				voice->roughness = value;
			break;
		case V_CLARITY: // formantshape
			if (sscanf(p, "%d", &value) == 1) {
				if (value > 4) {
					voice->peak_shape = 1; // squarer formant peaks
					value = 4;
				}
				voice->n_harmonic_peaks = 1+value;
			}
			break;
		case V_TONE:
		{
			int tone_data[12];
			ReadTonePoints(p, tone_data);
			SetToneAdjust(voice, tone_data);
		}
			break;
		case V_VOICING:
			if (sscanf(p, "%d", &value) == 1)
				voice->voicing = (value * 64)/100;
			break;
		case V_BREATH:
			voice->breath[0] = Read8Numbers(p, &voice->breath[1]);
			for (ix = 1; ix < 8; ix++) {
				if (ix % 2)
					voice->breath[ix] = -voice->breath[ix];
			}
			break;
		case V_BREATHW:
			voice->breathw[0] = Read8Numbers(p, &voice->breathw[1]);
			break;
		case V_CONSONANTS:
			value = sscanf(p, "%d %d", &voice->consonant_amp, &voice->consonant_ampv);
			break;
		case V_SPEED:
			sscanf(p, "%d", &voice->speed_percent);
			break;
		case V_MBROLA:
		{
			int srate = 16000;

			name2[0] = 0;
			sscanf(p, "%s %s %d", name1, name2, &srate);
			espeak_ng_STATUS status = LoadMbrolaTable(name1, name2, &srate);
			if (status != ENS_OK)
				espeak_ng_PrintStatusCodeMessage(status, stderr, NULL);
			else
				voice->samplerate = srate;
		}
			break;
		case V_KLATT:
			voice->klattv[0] = 1; // default source: IMPULSIVE
			Read8Numbers(p, voice->klattv);
			voice->klattv[KLATT_Kopen] -= 40;
			break;
		case V_FAST:
			Read8Numbers(p, speed.fast_settings);
			SetSpeed(3);
			break;
		case V_DICTMIN:
			sscanf(p, "%d", &dict_min);
			break;
		case V_MAINTAINER:
		case V_STATUS:
			break;

		case V_LETTER_VOWEL: {
			char str[5] = "";
			char c = '0';
			char *endptr = NULL;
			sscanf(p, "%s", str);
			// assume a hex value if string starts with "0x"
			if (str[0] == '0' && str[1] == 'x') {
				c = strtoul(str, &endptr, 16);
				if (errno == ERANGE)
					fprintf(stderr, "letterVowel out of range.\n");
			}
			else  {// otherwise, assume a single letter
				c = str[0];
				if (c < 97 || c > 122) // valid values are a-z, ascii 97-122
					fprintf(stderr, "letterVowel out of range.\n");
			}
 			new_translator->letter_bits[c] = (new_translator->letter_bits[c] & 0x40) | 0x81; // keep value for group 6 (front vowels e,i,y)
			break;
			}

		default:
			if ((key & 0xff00) == 0x100) {
				if (langopts)
					sscanf(p, "%d", &langopts->param[key &0xff]);
				else
					fprintf(stderr, "Cannot set voice attribute: language not set, or is invalid.\n");
			} else
				fprintf(stderr, "Bad voice attribute: %s\n", buf);
			break;
		}
	}
	if (f_voice != NULL)
		fclose(f_voice);

	if ((new_translator == NULL) && (!tone_only)) {
		// not set by language attribute
		new_translator = SelectTranslator(translator_name);
	}

	SetSpeed(3); // for speed_percent

	for (ix = 0; ix < N_PEAKS; ix++) {
		voice->freq2[ix] = voice->freq[ix];
		voice->height2[ix] = voice->height[ix];
		voice->width2[ix] = voice->width[ix];
	}

	if (tone_only)
		new_translator = translator;
	else {
		if ((ix = SelectPhonemeTableName(phonemes_name)) < 0) {
			fprintf(stderr, "Unknown phoneme table: '%s'\n", phonemes_name);
			ix = 0;
		}
		voice->phoneme_tab_ix = ix;
		new_translator->phoneme_tab_ix = ix;
		new_translator->dict_min_size = dict_min;
		LoadDictionary(new_translator, new_dictionary, control & 4);
		if (dictionary_name[0] == 0) {
			DeleteTranslator(new_translator);
			return NULL; // no dictionary loaded
		}

		new_translator->dict_condition = conditional_rules;

		voice_languages[langix] = 0;
	}

	langopts = &new_translator->langopts;

	if ((value = langopts->param[LOPT_LENGTH_MODS]) != 0)
		SetLengthMods(new_translator, value);

	voice->width[0] = (voice->width[0] * 105)/100;

	if (!tone_only)
		translator = new_translator;

	// relative lengths of different stress syllables
	for (ix = 0; ix < stress_lengths_set; ix++)
		translator->stress_lengths[ix] = stress_lengths[ix];
	for (ix = 0; ix < stress_add_set; ix++)
		translator->stress_lengths[ix] += stress_add[ix];
	for (ix = 0; ix < stress_amps_set; ix++) {
		translator->stress_amps[ix] = stress_amps[ix];
		translator->stress_amps_r[ix] = stress_amps[ix] -1;
	}

	return voice;
}

static char *ExtractVoiceVariantName(char *vname, int variant_num, int add_dir)
{
	// Remove any voice variant suffix (name or number) from a voice name
	// Returns the voice variant name

	char *p;
	static char variant_name[40];
	char variant_prefix[5];

	variant_name[0] = 0;
	sprintf(variant_prefix, "!v%c", PATHSEP);
	if (add_dir == 0)
		variant_prefix[0] = 0;

	if (vname != NULL) {
		if ((p = strchr(vname, '+')) != NULL) {
			// The voice name has a +variant suffix
			variant_num = 0;
			*p++ = 0; // delete the suffix from the voice name
			if (IsDigit09(*p))
				variant_num = atoi(p); // variant number
			else {
				// voice variant name, not number
				sprintf(variant_name, "%s%s", variant_prefix, p);
			}
		}
	}

	if (variant_num > 0) {
		if (variant_num < 10)
			sprintf(variant_name, "%sm%d", variant_prefix, variant_num); // male
		else
			sprintf(variant_name, "%sf%d", variant_prefix, variant_num-10); // female
	}

	return variant_name;
}

voice_t *LoadVoiceVariant(const char *vname, int variant_num)
{
	// Load a voice file.
	// Also apply a voice variant if specified by "variant", or by "+number" or "+name" in the "vname"

	voice_t *v;
	char *variant_name;
	char buf[60];

	strncpy0(buf, vname, sizeof(buf));
	variant_name = ExtractVoiceVariantName(buf, variant_num, 1);

	if ((v = LoadVoice(buf, 0)) == NULL)
		return NULL;

	if (variant_name[0] != 0)
		v = LoadVoice(variant_name, 2);
	return v;
}

static int __cdecl VoiceNameSorter(const void *p1, const void *p2)
{
	int ix;
	espeak_VOICE *v1 = *(espeak_VOICE **)p1;
	espeak_VOICE *v2 = *(espeak_VOICE **)p2;

	if ((ix = strcmp(&v1->languages[1], &v2->languages[1])) != 0) // primary language name
		return ix;
	if ((ix = v1->languages[0] - v2->languages[0]) != 0) // priority number
		return ix;
	return strcmp(v1->name, v2->name);
}

static int __cdecl VoiceScoreSorter(const void *p1, const void *p2)
{
	int ix;
	espeak_VOICE *v1 = *(espeak_VOICE **)p1;
	espeak_VOICE *v2 = *(espeak_VOICE **)p2;

	if ((ix = v2->score - v1->score) != 0)
		return ix;
	return strcmp(v1->name, v2->name);
}

static int ScoreVoice(espeak_VOICE *voice_spec, const char *spec_language, int spec_n_parts, int spec_lang_len, espeak_VOICE *voice)
{
	int ix;
	const char *p;
	int c1, c2;
	int language_priority;
	int n_parts;
	int matching;
	int matching_parts;
	int score = 0;
	int x;
	int ratio;
	int required_age;
	int diff;

	p = voice->languages; // list of languages+dialects for which this voice is suitable

	if (spec_n_parts < 0) {
		// match on the subdirectory
		if (memcmp(voice->identifier, spec_language, spec_lang_len) == 0)
			return 100;
		return 0;
	}

	if (spec_n_parts == 0)
		score = 100;
	else {
		if ((*p == 0) && (strcmp(spec_language, "variants") == 0)) {
			// match on a voice with no languages if the required language is "variants"
			score = 100;
		}

		// compare the required language with each of the languages of this voice
		while (*p != 0) {
			language_priority = *p++;

			matching = 1;
			matching_parts = 0;
			n_parts = 1;

			for (ix = 0;; ix++) {
				if ((ix >= spec_lang_len) || ((c1 = spec_language[ix]) == '-'))
					c1 = 0;
				if ((c2 = p[ix]) == '-')
					c2 = 0;

				if (c1 != c2)
					matching = 0;

				if (p[ix] == '-') {
					n_parts++;
					if (matching)
						matching_parts++;
				}
				if (p[ix] == 0)
					break;
			}
			p += (ix+1);
			matching_parts += matching; // number of parts which match

			if (matching_parts == 0)
				continue; // no matching parts for this language

			x = 5;
			// reduce the score if not all parts of the required language match
			if ((diff = (spec_n_parts - matching_parts)) > 0)
				x -= diff;

			// reduce score if the language is more specific than required
			if ((diff = (n_parts - matching_parts)) > 0)
				x -= diff;

			x = x*100 - (language_priority * 2);

			if (x > score)
				score = x;
		}
	}
	if (score == 0)
		return 0;

	if (voice_spec->name != NULL) {
		if (strcmp(voice_spec->name, voice->name) == 0) {
			// match on voice name
			score += 500;
		} else if (strcmp(voice_spec->name, voice->identifier) == 0)
			score += 400;
	}

	if (((voice_spec->gender == ENGENDER_MALE) || (voice_spec->gender == ENGENDER_FEMALE)) &&
	    ((voice->gender == ENGENDER_MALE) || (voice->gender == ENGENDER_FEMALE))) {
		if (voice_spec->gender == voice->gender)
			score += 50;
		else
			score -= 50;
	}

	if ((voice_spec->age <= 12) && (voice->gender == ENGENDER_FEMALE) && (voice->age > 12))
		score += 5; // give some preference for non-child female voice if a child is requested

	if (voice->age != 0) {
		if (voice_spec->age == 0)
			required_age = 30;
		else
			required_age = voice_spec->age;

		ratio = (required_age*100)/voice->age;
		if (ratio < 100)
			ratio = 10000/ratio;
		ratio = (ratio - 100)/10; // 0=exact match, 10=out by factor of 2
		x = 5 - ratio;
		if (x > 0) x = 0;

		score = score + x;

		if (voice_spec->age > 0)
			score += 10; // required age specified, favour voices with a specified age (near it)
	}
	if (score < 1)
		score = 1;
	return score;
}

static int SetVoiceScores(espeak_VOICE *voice_select, espeak_VOICE **voices, int control)
{
	// control: bit0=1  include mbrola voices
	int ix;
	int score;
	int nv; // number of candidates
	int n_parts = 0;
	int lang_len = 0;
	espeak_VOICE *vp;
	char language[80];
	char buf[sizeof(path_home)+80];

	// count number of parts in the specified language
	if ((voice_select->languages != NULL) && (voice_select->languages[0] != 0)) {
		n_parts = 1;
		lang_len = strlen(voice_select->languages);
		for (ix = 0; (ix <= lang_len) && ((unsigned)ix < sizeof(language)); ix++) {
			if ((language[ix] = tolower(voice_select->languages[ix])) == '-')
				n_parts++;
		}
	}

	if ((n_parts == 1) && (control & 1)) {
		if (strcmp(language, "mbrola") == 0) {
			language[2] = 0; // truncate to "mb"
			lang_len = 2;
		}

		sprintf(buf, "%s/voices/%s", path_home, language);
		if (GetFileLength(buf) == -EISDIR) {
			// A subdirectory name has been specified.  List all the voices in that subdirectory
			language[lang_len++] = PATHSEP;
			language[lang_len] = 0;
			n_parts = -1;
		}
	}

	// select those voices which match the specified language
	nv = 0;
	for (ix = 0; ix < n_voices_list; ix++) {
		vp = voices_list[ix];

		if (((control & 1) == 0) && (memcmp(vp->identifier, "mb/", 3) == 0))
			continue;

		if ((score = ScoreVoice(voice_select, language, n_parts, lang_len, voices_list[ix])) > 0) {
			voices[nv++] = vp;
			vp->score = score;
		}
	}
	voices[nv] = NULL; // list terminator

	if (nv == 0)
		return 0;

	// sort the selected voices by their score
	qsort(voices, nv, sizeof(espeak_VOICE *), (int(__cdecl *)(const void *, const void *))VoiceScoreSorter);

	return nv;
}

espeak_VOICE *SelectVoiceByName(espeak_VOICE **voices, const char *name2)
{
	int ix;
	int match_fname = -1;
	int match_fname2 = -1;
	int match_name = -1;
	const char *id; // this is the filename within espeak-ng-data/voices
	char *variant_name;
	int last_part_len;
	char last_part[41];
	char name[40];

	if (voices == NULL) {
		if (n_voices_list == 0)
			espeak_ListVoices(NULL); // create the voices list
		voices = voices_list;
	}

	strncpy0(name, name2, sizeof(name));
	if ((variant_name = strchr(name, '+')) != NULL) {
		*variant_name = 0;
		variant_name++;
	}

	sprintf(last_part, "%c%s", PATHSEP, name);
	last_part_len = strlen(last_part);

	for (ix = 0; voices[ix] != NULL; ix++) {
		if (strcasecmp(name, voices[ix]->name) == 0) {
			match_name = ix; // found matching voice name
			break;
		} else {
			id = voices[ix]->identifier;
			if (strcasecmp(name, id) == 0)
				match_fname = ix; // matching identifier, use this if no matching name
			else if (strcasecmp(last_part, &id[strlen(id)-last_part_len]) == 0)
				match_fname2 = ix;
		}
	}

	if (match_name < 0) {
		match_name = match_fname; // no matching name, try matching filename
		if (match_name < 0)
			match_name = match_fname2; // try matching just the last part of the filename
	}

	if (match_name < 0)
		return NULL;

	return voices[match_name];
}

char const *SelectVoice(espeak_VOICE *voice_select, int *found)
{
	// Returns a path within espeak-voices, with a possible +variant suffix
	// variant is an output-only parameter

	int nv; // number of candidates
	int ix, ix2;
	int j;
	int n_variants;
	int variant_number;
	int gender;
	int skip;
	int aged = 1;
	char *variant_name;
	const char *p, *p_start;
	espeak_VOICE *vp = NULL;
	espeak_VOICE *vp2;
	espeak_VOICE voice_select2;
	espeak_VOICE *voices[N_VOICES_LIST]; // list of candidates
	espeak_VOICE *voices2[N_VOICES_LIST+N_VOICE_VARIANTS];
	static espeak_VOICE voice_variants[N_VOICE_VARIANTS];
	static char voice_id[50];

	*found = 1;
	memcpy(&voice_select2, voice_select, sizeof(voice_select2));

	if (n_voices_list == 0)
		espeak_ListVoices(NULL); // create the voices list

	if ((voice_select2.languages == NULL) || (voice_select2.languages[0] == 0)) {
		// no language is specified. Get language from the named voice
		static char buf[60];

		if (voice_select2.name == NULL) {
			if ((voice_select2.name = voice_select2.identifier) == NULL)
				voice_select2.name = ESPEAKNG_DEFAULT_VOICE;
		}

		strncpy0(buf, voice_select2.name, sizeof(buf));
		variant_name = ExtractVoiceVariantName(buf, 0, 0);

		vp = SelectVoiceByName(voices_list, buf);
		if (vp != NULL) {
			voice_select2.languages = &(vp->languages[1]);

			if ((voice_select2.gender == ENGENDER_UNKNOWN) && (voice_select2.age == 0) && (voice_select2.variant == 0)) {
				if (variant_name[0] != 0) {
					sprintf(voice_id, "%s+%s", vp->identifier, variant_name);
					return voice_id;
				}

				return vp->identifier;
			}
		}
	}

	// select and sort voices for the required language
	nv = SetVoiceScores(&voice_select2, voices, 0);

	if (nv == 0) {
		// no matching voice, choose the default
		*found = 0;
		if ((voices[0] = SelectVoiceByName(voices_list, ESPEAKNG_DEFAULT_VOICE)) != NULL)
			nv = 1;
	}

	gender = 0;
	if ((voice_select2.gender == ENGENDER_FEMALE) || ((voice_select2.age > 0) && (voice_select2.age < 13)))
		gender = ENGENDER_FEMALE;
	else if (voice_select2.gender == ENGENDER_MALE)
		gender = ENGENDER_MALE;

	#define AGE_OLD 60
	if (voice_select2.age < AGE_OLD)
		aged = 0;

	p = p_start = variant_lists[gender];
	if (aged == 0)
		p++; // the first voice in the variants list is older

	// add variants for the top voices
	n_variants = 0;
	for (ix = 0, ix2 = 0; ix < nv; ix++) {
		vp = voices[ix];
		// is the main voice the required gender?
		skip = 0;

		if ((gender != ENGENDER_UNKNOWN) && (vp->gender != gender))
			skip = 1;
		if ((ix2 == 0) && aged && (vp->age < AGE_OLD))
			skip = 1;

		if (skip == 0)
			voices2[ix2++] = vp;

		for (j = 0; (j < vp->xx1) && (n_variants < N_VOICE_VARIANTS);) {
			if ((variant_number = *p) == 0) {
				p = p_start;
				continue;
			}

			vp2 = &voice_variants[n_variants++]; // allocate space for voice variant
			memcpy(vp2, vp, sizeof(espeak_VOICE)); // copy from the original voice
			vp2->variant = variant_number;
			voices2[ix2++] = vp2;
			p++;
			j++;
		}
	}
	// add any more variants to the end of the list
	while ((vp != NULL) && ((variant_number = *p++) != 0) && (n_variants < N_VOICE_VARIANTS)) {
		vp2 = &voice_variants[n_variants++]; // allocate space for voice variant
		memcpy(vp2, vp, sizeof(espeak_VOICE)); // copy from the original voice
		vp2->variant = variant_number;
		voices2[ix2++] = vp2;
	}

	// index the sorted list by the required variant number
	if (ix2 == 0)
		return NULL;
	vp = voices2[voice_select2.variant % ix2];

	if (vp->variant != 0) {
		variant_name = ExtractVoiceVariantName(NULL, vp->variant, 0);
		sprintf(voice_id, "%s+%s", vp->identifier, variant_name);
		return voice_id;
	}

	return vp->identifier;
}

static void GetVoices(const char *path, int len_path_voices, int is_language_file)
{
	FILE *f_voice;
	espeak_VOICE *voice_data;
	int ftype;
	char fname[sizeof(path_home)+100];

#ifdef PLATFORM_WINDOWS
	WIN32_FIND_DATAA FindFileData;
	HANDLE hFind = INVALID_HANDLE_VALUE;

	#undef UNICODE // we need FindFirstFileA() which takes an 8-bit c-string
	sprintf(fname, "%s\\*", path);
	hFind = FindFirstFileA(fname, &FindFileData);
	if (hFind == INVALID_HANDLE_VALUE)
		return;

	do {
		if (n_voices_list >= (N_VOICES_LIST-2))
			break; // voices list is full

		if (FindFileData.cFileName[0] != '.') {
			sprintf(fname, "%s%c%s", path, PATHSEP, FindFileData.cFileName);
			ftype = GetFileLength(fname);

			if (ftype == -EISDIR) {
				// a sub-directory
				GetVoices(fname, len_path_voices, is_language_file);
			} else if (ftype > 0) {
				// a regular file, add it to the voices list
				if ((f_voice = fopen(fname, "r")) == NULL)
					continue;

				// pass voice file name within the voices directory
				voice_data = ReadVoiceFile(f_voice, fname+len_path_voices, is_language_file);
				fclose(f_voice);

				if (voice_data != NULL)
					voices_list[n_voices_list++] = voice_data;
			}
		}
	} while (FindNextFileA(hFind, &FindFileData) != 0);
	FindClose(hFind);
#else
	DIR *dir;
	struct dirent *ent;

	if ((dir = opendir((char *)path)) == NULL) // note: (char *) is needed for WINCE
		return;

	while ((ent = readdir(dir)) != NULL) {
		if (n_voices_list >= (N_VOICES_LIST-2))
			break; // voices list is full

		if (ent->d_name[0] == '.')
			continue;

		sprintf(fname, "%s%c%s", path, PATHSEP, ent->d_name);

		ftype = GetFileLength(fname);

		if (ftype == -EISDIR) {
			// a sub-directory
			GetVoices(fname, len_path_voices, is_language_file);
		} else if (ftype > 0) {
			// a regular file, add it to the voices list
			if ((f_voice = fopen(fname, "r")) == NULL)
				continue;

			// pass voice file name within the voices directory
			voice_data = ReadVoiceFile(f_voice, fname+len_path_voices, is_language_file);
			fclose(f_voice);

			if (voice_data != NULL)
				voices_list[n_voices_list++] = voice_data;
		}
	}
	closedir(dir);
#endif
}

#pragma GCC visibility push(default)

ESPEAK_NG_API espeak_ng_STATUS espeak_ng_SetVoiceByName(const char *name)
{
	espeak_VOICE *v;
	int ix;
	espeak_VOICE voice_selector;
	char *variant_name;
	static char buf[60];

	strncpy0(buf, name, sizeof(buf));

	variant_name = ExtractVoiceVariantName(buf, 0, 1);

	for (ix = 0;; ix++) {
		// convert voice name to lower case  (ascii)
		if ((buf[ix] = tolower(buf[ix])) == 0)
			break;
	}

	memset(&voice_selector, 0, sizeof(voice_selector));
	voice_selector.name = (char *)name; // include variant name in voice stack ??

	// first check for a voice with this filename
	// This may avoid the need to call espeak_ListVoices().

	if (LoadVoice(buf, 1) != NULL) {
		if (variant_name[0] != 0)
			LoadVoice(variant_name, 2);

		DoVoiceChange(voice);
		voice_selector.languages = voice->language_name;
		SetVoiceStack(&voice_selector, variant_name);
		return ENS_OK;
	}

	if (n_voices_list == 0)
		espeak_ListVoices(NULL); // create the voices list

	if ((v = SelectVoiceByName(voices_list, buf)) != NULL) {
		if (LoadVoice(v->identifier, 0) != NULL) {
			if (variant_name[0] != 0)
				LoadVoice(variant_name, 2);
			DoVoiceChange(voice);
			voice_selector.languages = voice->language_name;
			SetVoiceStack(&voice_selector, variant_name);
			return ENS_OK;
		}
	}
	return ENS_VOICE_NOT_FOUND;
}

ESPEAK_NG_API espeak_ng_STATUS espeak_ng_SetVoiceByProperties(espeak_VOICE *voice_selector)
{
	const char *voice_id;
	int voice_found;

	voice_id = SelectVoice(voice_selector, &voice_found);
	if (voice_found == 0)
		return ENS_VOICE_NOT_FOUND;

	LoadVoiceVariant(voice_id, 0);
	DoVoiceChange(voice);
	SetVoiceStack(voice_selector, "");

	return ENS_OK;
}

#pragma GCC visibility pop

void FreeVoiceList()
{
	int ix;
	for (ix = 0; ix < n_voices_list; ix++) {
		if (voices_list[ix] != NULL) {
			free(voices_list[ix]);
			voices_list[ix] = NULL;
		}
	}
	n_voices_list = 0;
}

#pragma GCC visibility push(default)

ESPEAK_API const espeak_VOICE **espeak_ListVoices(espeak_VOICE *voice_spec)
{
	char path_voices[sizeof(path_home)+12];

	int ix;
	int j;
	espeak_VOICE *v;
	static espeak_VOICE **voices = NULL;

	// free previous voice list data
	FreeVoiceList();

	sprintf(path_voices, "%s%cvoices", path_home, PATHSEP);
	GetVoices(path_voices, strlen(path_voices)+1, 0);

	sprintf(path_voices, "%s%clang", path_home, PATHSEP);
	GetVoices(path_voices, strlen(path_voices)+1, 1);

	voices_list[n_voices_list] = NULL; // voices list terminator
	espeak_VOICE **new_voices = (espeak_VOICE **)realloc(voices, sizeof(espeak_VOICE *)*(n_voices_list+1));
	if (new_voices == NULL)
		return (const espeak_VOICE **)voices;
	voices = new_voices;

	// sort the voices list
	qsort(voices_list, n_voices_list, sizeof(espeak_VOICE *),
	      (int(__cdecl *)(const void *, const void *))VoiceNameSorter);

	if (voice_spec) {
		// select the voices which match the voice_spec, and sort them by preference
		SetVoiceScores(voice_spec, voices, 1);
	} else {
		// list all: omit variant and mbrola voices
		j = 0;
		for (ix = 0; (v = voices_list[ix]) != NULL; ix++) {
			if ((v->languages[0] != 0) && (strcmp(&v->languages[1], "variant") != 0)
			    && (memcmp(v->identifier, "mb/", 3) != 0))
				voices[j++] = v;
		}
		voices[j] = NULL;
	}
	return (const espeak_VOICE **)voices;
}

ESPEAK_API espeak_VOICE *espeak_GetCurrentVoice(void)
{
	return &current_voice_selected;
}

#pragma GCC visibility pop