1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192939495969798991001011021031041051061071081091101111121131141151161171181191201211221231241251261271281291301311321331341351361371381391401411421431441451461471481491501511521531541551561571581591601611621631641651661671681691701711721731741751761771781791801811821831841851861871881891901911921931941951961971981992002012022032042052062072082092102112122132142152162172182192202212222232242252262272282292302312322332342352362372382392402412422432442452462472482492502512522532542552562572582592602612622632642652662672682692702712722732742752762772782792802812822832842852862872882892902912922932942952962972982993003013023033043053063073083093103113123133143153163173183193203213223233243253263273283293303313323333343353363373383393403413423433443453463473483493503513523533543553563573583593603613623633643653663673683693703713723733743753763773783793803813823833843853863873883893903913923933943953963973983994004014024034044054064074084094104114124134144154164174184194204214224234244254264274284294304314324334344354364374384394404414424434444454464474484494504514524534544554564574584594604614624634644654664674684694704714724734744754764774784794804814824834844854864874884894904914924934944954964974984995005015025035045055065075085095105115125135145155165175185195205215225235245255265275285295305315325335345355365375385395405415425435445455465475485495505515525535545555565575585595605615625635645655665675685695705715725735745755765775785795805815825835845855865875885895905915925935945955965975985996006016026036046056066076086096106116126136146156166176186196206216226236246256266276286296306316326336346356366376386396406416426436446456466476486496506516526536546556566576586596606616626636646656666676686696706716726736746756766776786796806816826836846856866876886896906916926936946956966976986997007017027037047057067077087097107117127137147157167177187197207217227237247257267277287297307317327337347357367377387397407417427437447457467477487497507517527537547557567577587597607617627637647657667677687697707717727737747757767777787797807817827837847857867877887897907917927937947957967977987998008018028038048058068078088098108118128138148158168178188198208218228238248258268278288298308318328338348358368378388398408418428438448458468478488498508518528538548558568578588598608618628638648658668678688698708718728738748758768778788798808818828838848858868878888898908918928938948958968978988999009019029039049059069079089099109119129139149159169179189199209219229239249259269279289299309319329339349359369379389399409419429439449459469479489499509519529539549559569579589599609619629639649659669679689699709719729739749759769779789799809819829839849859869879889899909919929939949959969979989991000100110021003100410051006100710081009101010111012101310141015101610171018101910201021102210231024102510261027102810291030103110321033103410351036103710381039104010411042104310441045104610471048104910501051105210531054105510561057105810591060106110621063106410651066106710681069107010711072107310741075107610771078107910801081108210831084108510861087108810891090109110921093109410951096109710981099110011011102110311041105110611071108110911101111111211131114111511161117111811191120112111221123112411251126112711281129113011311132113311341135113611371138113911401141114211431144114511461147114811491150115111521153115411551156115711581159116011611162116311641165116611671168116911701171117211731174117511761177117811791180118111821183118411851186118711881189119011911192119311941195119611971198119912001201120212031204120512061207120812091210121112121213121412151216121712181219122012211222122312241225122612271228122912301231123212331234123512361237123812391240124112421243124412451246124712481249125012511252125312541255125612571258125912601261126212631264126512661267126812691270127112721273127412751276127712781279128012811282128312841285128612871288128912901291129212931294129512961297129812991300130113021303130413051306130713081309131013111312131313141315131613171318131913201321132213231324132513261327132813291330133113321333133413351336133713381339134013411342134313441345134613471348134913501351135213531354135513561357135813591360136113621363136413651366136713681369137013711372137313741375137613771378137913801381138213831384138513861387138813891390139113921393139413951396139713981399140014011402140314041405140614071408140914101411141214131414141514161417141814191420142114221423142414251426142714281429143014311432143314341435143614371438143914401441144214431444144514461447144814491450145114521453145414551456145714581459146014611462146314641465146614671468146914701471147214731474147514761477147814791480148114821483148414851486148714881489149014911492149314941495149614971498149915001501150215031504150515061507150815091510151115121513151415151516151715181519152015211522152315241525152615271528152915301531153215331534153515361537153815391540154115421543154415451546154715481549155015511552155315541555155615571558155915601561156215631564156515661567156815691570 |
- /*
- * Copyright (C) 2005 to 2015 by Jonathan Duddington
- * email: jonsd@users.sourceforge.net
- * Copyright (C) 2015-2016 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 <locale.h>
- #include <stdint.h>
- #include <stdio.h>
- #include <stdlib.h>
- #include <string.h>
- #include "espeak_ng.h"
- #include "speak_lib.h"
- #include "encoding.h"
- #include "speech.h"
- #include "synthesize.h"
- #include "translate.h"
- // start of unicode pages for character sets
- #define OFFSET_GREEK 0x380
- #define OFFSET_CYRILLIC 0x420
- #define OFFSET_ARMENIAN 0x530
- #define OFFSET_HEBREW 0x590
- #define OFFSET_ARABIC 0x600
- #define OFFSET_SYRIAC 0x700
- #define OFFSET_THAANA 0x780 // Divehi/Maldives
- #define OFFSET_DEVANAGARI 0x900
- #define OFFSET_BENGALI 0x980
- #define OFFSET_GURMUKHI 0xa00
- #define OFFSET_GUJARATI 0xa80
- #define OFFSET_ORIYA 0xb00
- #define OFFSET_TAMIL 0xb80
- #define OFFSET_TELUGU 0xc00
- #define OFFSET_KANNADA 0xc80
- #define OFFSET_MALAYALAM 0xd00
- #define OFFSET_SINHALA 0x0d80
- #define OFFSET_THAI 0x0e00
- #define OFFSET_LAO 0x0e80
- #define OFFSET_TIBET 0x0f00
- #define OFFSET_MYANMAR 0x1000
- #define OFFSET_GEORGIAN 0x10a0
- #define OFFSET_KOREAN 0x1100
- #define OFFSET_ETHIOPIC 0x1200
- // character ranges must be listed in ascending unicode order
- ALPHABET alphabets[] = {
- { "_el", OFFSET_GREEK, 0x380, 0x3ff, L('e', 'l'), AL_DONT_NAME | AL_NOT_LETTERS | AL_WORDS },
- { "_cyr", OFFSET_CYRILLIC, 0x400, 0x52f, 0, 0 },
- { "_hy", OFFSET_ARMENIAN, 0x530, 0x58f, L('h', 'y'), AL_WORDS },
- { "_he", OFFSET_HEBREW, 0x590, 0x5ff, 0, 0 },
- { "_ar", OFFSET_ARABIC, 0x600, 0x6ff, 0, 0 },
- { "_syc", OFFSET_SYRIAC, 0x700, 0x74f, 0, 0 },
- { "_dv", OFFSET_THAANA, 0x780, 0x7bf, 0, 0 },
- { "_hi", OFFSET_DEVANAGARI, 0x900, 0x97f, L('h', 'i'), AL_WORDS },
- { "_bn", OFFSET_BENGALI, 0x0980, 0x9ff, L('b', 'n'), AL_WORDS },
- { "_gur", OFFSET_GURMUKHI, 0xa00, 0xa7f, L('p', 'a'), AL_WORDS },
- { "_gu", OFFSET_GUJARATI, 0xa80, 0xaff, L('g', 'u'), AL_WORDS },
- { "_or", OFFSET_ORIYA, 0xb00, 0xb7f, 0, 0 },
- { "_ta", OFFSET_TAMIL, 0xb80, 0xbff, L('t', 'a'), AL_WORDS },
- { "_te", OFFSET_TELUGU, 0xc00, 0xc7f, L('t', 'e'), 0 },
- { "_kn", OFFSET_KANNADA, 0xc80, 0xcff, L('k', 'n'), AL_WORDS },
- { "_ml", OFFSET_MALAYALAM, 0xd00, 0xd7f, L('m', 'l'), AL_WORDS },
- { "_si", OFFSET_SINHALA, 0xd80, 0xdff, L('s', 'i'), AL_WORDS },
- { "_th", OFFSET_THAI, 0xe00, 0xe7f, 0, 0 },
- { "_lo", OFFSET_LAO, 0xe80, 0xeff, 0, 0 },
- { "_ti", OFFSET_TIBET, 0xf00, 0xfff, 0, 0 },
- { "_my", OFFSET_MYANMAR, 0x1000, 0x109f, 0, 0 },
- { "_ka", OFFSET_GEORGIAN, 0x10a0, 0x10ff, L('k', 'a'), AL_WORDS },
- { "_ko", OFFSET_KOREAN, 0x1100, 0x11ff, L('k', 'o'), AL_WORDS },
- { "_eth", OFFSET_ETHIOPIC, 0x1200, 0x139f, 0, 0 },
- { "_braille", 0x2800, 0x2800, 0x28ff, 0, AL_NO_SYMBOL },
- { "_ja", 0x3040, 0x3040, 0x30ff, 0, AL_NOT_CODE },
- { "_zh", 0x3100, 0x3100, 0x9fff, 0, AL_NOT_CODE },
- { "_ko", 0xa700, 0xa700, 0xd7ff, L('k', 'o'), AL_NOT_CODE | AL_WORDS },
- { NULL, 0, 0, 0, 0, 0 }
- };
- ALPHABET *AlphabetFromChar(int c)
- {
- // Find the alphabet from a character.
- ALPHABET *alphabet = alphabets;
- while (alphabet->name != NULL) {
- if (c <= alphabet->range_max) {
- if (c >= alphabet->range_min)
- return alphabet;
- else
- break;
- }
- alphabet++;
- }
- return NULL;
- }
- static void Translator_Russian(Translator *tr);
- static void ResetLetterBits(Translator *tr, int groups)
- {
- // Clear all the specified groups
- unsigned int ix;
- unsigned int mask;
- mask = ~groups;
- for (ix = 0; ix < sizeof(tr->letter_bits); ix++)
- tr->letter_bits[ix] &= mask;
- }
- static void SetLetterBits(Translator *tr, int group, const char *string)
- {
- int bits;
- unsigned char c;
- bits = (1L << group);
- while ((c = *string++) != 0)
- tr->letter_bits[c] |= bits;
- }
- static void SetLetterBitsRange(Translator *tr, int group, int first, int last)
- {
- int bits;
- int ix;
- bits = (1L << group);
- for (ix = first; ix <= last; ix++)
- tr->letter_bits[ix] |= bits;
- }
- // ignore these characters
- static const unsigned short chars_ignore_default[] = {
- // U+00AD SOFT HYPHEN
- // Used to mark hyphenation points in words for where to split a
- // word at the end of a line to provide readable justified text.
- 0xad, 1,
- // U+200C ZERO WIDTH NON-JOINER
- // Used to prevent combined ligatures being displayed in their
- // combined form.
- 0x200c, 1,
- // U+200D ZERO WIDTH JOINER
- // Used to indicate an alternative connected form made up of the
- // characters surrounding the ZWJ in Devanagari, Kannada, Malayalam
- // and Emoji.
- // 0x200d, 1, // Not ignored.
- // End of the ignored character list.
- 0, 0
- };
- // alternatively, ignore characters but allow zero-width-non-joiner (lang-fa)
- static const unsigned short chars_ignore_zwnj_hyphen[] = {
- // U+00AD SOFT HYPHEN
- // Used to mark hyphenation points in words for where to split a
- // word at the end of a line to provide readable justified text.
- 0xad, 1,
- // U+0640 TATWEEL (KASHIDA)
- // Used in Arabic scripts to stretch characters for justifying
- // the text.
- 0x640, 1,
- // U+200C ZERO WIDTH NON-JOINER
- // Used to prevent combined ligatures being displayed in their
- // combined form.
- 0x200c, '-',
- // U+200D ZERO WIDTH JOINER
- // Used to indicate an alternative connected form made up of the
- // characters surrounding the ZWJ in Devanagari, Kannada, Malayalam
- // and Emoji.
- // 0x200d, 1, // Not ignored.
- // End of the ignored character list.
- 0, 0
- };
- const unsigned char utf8_ordinal[] = { 0xc2, 0xba, 0 }; // masculine ordinal character, UTF-8
- const unsigned char utf8_null[] = { 0 }; // null string, UTF-8
- static Translator *NewTranslator(void)
- {
- Translator *tr;
- int ix;
- static const unsigned char stress_amps2[] = { 18, 18, 20, 20, 20, 22, 22, 20 };
- static const short stress_lengths2[8] = { 182, 140, 220, 220, 220, 240, 260, 280 };
- static const wchar_t empty_wstring[1] = { 0 };
- static const wchar_t punct_in_word[2] = { '\'', 0 }; // allow hyphen within words
- static const unsigned char default_tunes[6] = { 0, 1, 2, 3, 0, 0 };
- // Translates character codes in the range transpose_min to transpose_max to
- // a number in the range 1 to 63. 0 indicates there is no translation.
- // Used up to 57 (max of 63)
- static const char transpose_map_latin[] = {
- 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, // 0x60
- 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 0, 0, 0, 0, 0, // 0x70
- 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, // 0x80
- 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, // 0x90
- 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, // 0xa0
- 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, // 0xb0
- 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, // 0xc0
- 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, // 0xd0
- 27, 28, 29, 0, 0, 30, 31, 32, 33, 34, 35, 36, 0, 37, 38, 0, // 0xe0
- 0, 0, 0, 39, 0, 0, 40, 0, 41, 0, 42, 0, 43, 0, 0, 0, // 0xf0
- 0, 0, 0, 44, 0, 45, 0, 46, 0, 0, 0, 0, 0, 47, 0, 0, // 0x100
- 0, 48, 0, 0, 0, 0, 0, 0, 0, 49, 0, 0, 0, 0, 0, 0, // 0x110
- 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, // 0x120
- 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, // 0x130
- 0, 0, 50, 0, 51, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, // 0x140
- 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 52, 0, 0, 0, 0, // 0x150
- 0, 53, 0, 54, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, // 0x160
- 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 55, 0, 56, 0, 57, 0, // 0x170
- };
- if ((tr = (Translator *)malloc(sizeof(Translator))) == NULL)
- return NULL;
- tr->encoding = ESPEAKNG_ENCODING_ISO_8859_1;
- dictionary_name[0] = 0;
- tr->dictionary_name[0] = 0;
- tr->dict_condition = 0;
- tr->dict_min_size = 0;
- tr->data_dictrules = NULL; // language_1 translation rules file
- tr->data_dictlist = NULL; // language_2 dictionary lookup file
- tr->transpose_min = 0x60;
- tr->transpose_max = 0x17f;
- tr->transpose_map = transpose_map_latin;
- tr->frequent_pairs = NULL;
- // only need lower case
- tr->letter_bits_offset = 0;
- memset(tr->letter_bits, 0, sizeof(tr->letter_bits));
- memset(tr->letter_groups, 0, sizeof(tr->letter_groups));
- // 0-6 sets of characters matched by A B C H F G Y in pronunciation rules
- // these may be set differently for different languages
- SetLetterBits(tr, 0, "aeiou"); // A vowels, except y
- SetLetterBits(tr, 1, "bcdfgjklmnpqstvxz"); // B hard consonants, excluding h,r,w
- SetLetterBits(tr, 2, "bcdfghjklmnpqrstvwxz"); // C all consonants
- SetLetterBits(tr, 3, "hlmnr"); // H 'soft' consonants
- SetLetterBits(tr, 4, "cfhkpqstx"); // F voiceless consonants
- SetLetterBits(tr, 5, "bdgjlmnrvwyz"); // G voiced
- SetLetterBits(tr, 6, "eiy"); // Letter group Y, front vowels
- SetLetterBits(tr, 7, "aeiouy"); // vowels, including y
- tr->char_plus_apostrophe = empty_wstring;
- tr->punct_within_word = punct_in_word;
- tr->chars_ignore = chars_ignore_default;
- for (ix = 0; ix < 8; ix++) {
- tr->stress_amps[ix] = stress_amps2[ix];
- tr->stress_amps_r[ix] = stress_amps2[ix] - 1;
- tr->stress_lengths[ix] = stress_lengths2[ix];
- }
- memset(&(tr->langopts), 0, sizeof(tr->langopts));
- tr->langopts.max_lengthmod = 500;
- tr->langopts.lengthen_tonic = 20;
- tr->langopts.stress_rule = STRESSPOSN_2R;
- tr->langopts.unstressed_wd1 = 1;
- tr->langopts.unstressed_wd2 = 3;
- tr->langopts.param[LOPT_SONORANT_MIN] = 95;
- tr->langopts.param[LOPT_LONG_VOWEL_THRESHOLD] = 190/2;
- tr->langopts.param[LOPT_MAXAMP_EOC] = 19;
- tr->langopts.param[LOPT_UNPRONOUNCABLE] = 's'; // don't count this character at start of word
- tr->langopts.param[LOPT_BRACKET_PAUSE] = 4; // pause at bracket
- tr->langopts.param2[LOPT_BRACKET_PAUSE] = 2; // pauses when announcing bracket names
- tr->langopts.max_initial_consonants = 3;
- tr->langopts.replace_chars = NULL;
- tr->langopts.alt_alphabet_lang = L('e', 'n');
- tr->langopts.roman_suffix = utf8_null;
- SetLengthMods(tr, 201);
- tr->langopts.long_stop = 100;
- tr->langopts.max_roman = 49;
- tr->langopts.min_roman = 2;
- tr->langopts.thousands_sep = ',';
- tr->langopts.decimal_sep = '.';
- tr->langopts.break_numbers = BREAK_THOUSANDS; // 1000, 1000,000 1,000,000 etc
- tr->langopts.max_digits = 14;
- memcpy(tr->punct_to_tone, punctuation_to_tone, sizeof(tr->punct_to_tone));
- memcpy(tr->langopts.tunes, default_tunes, sizeof(tr->langopts.tunes));
- return tr;
- }
- // common letter pairs, encode these as a single byte
- // 2 bytes, using the transposed character codes
- static const short pairs_ru[] = {
- 0x010c, // ла 21052 0x23
- 0x010e, // на 18400
- 0x0113, // та 14254
- 0x0301, // ав 31083
- 0x030f, // ов 13420
- 0x060e, // не 21798
- 0x0611, // ре 19458
- 0x0903, // ви 16226
- 0x0b01, // ак 14456
- 0x0b0f, // ок 17836
- 0x0c01, // ал 13324
- 0x0c09, // ил 16877
- 0x0e01, // ан 15359
- 0x0e06, // ен 13543 0x30
- 0x0e09, // ин 17168
- 0x0e0e, // нн 15973
- 0x0e0f, // он 22373
- 0x0e1c, // ын 15052
- 0x0f03, // во 24947
- 0x0f11, // ро 13552
- 0x0f12, // со 16368
- 0x100f, // оп 19054
- 0x1011, // рп 17067
- 0x1101, // ар 23967
- 0x1106, // ер 18795
- 0x1109, // ир 13797
- 0x110f, // ор 21737
- 0x1213, // тс 25076
- 0x1220, // яс 14310
- 0x7fff
- };
- static const unsigned int replace_cyrillic_latin[] =
- { 0x430, 'a',
- 0x431, 'b',
- 0x446, 'c',
- 0x45b, 0x107,
- 0x447, 0x10d,
- 0x45f, 'd'+(0x17e<<16),
- 0x455, 'd'+('z'<<16),
- 0x434, 'd',
- 0x452, 0x111,
- 0x435, 'e',
- 0x444, 'f',
- 0x433, 'g',
- 0x445, 'h',
- 0x438, 'i',
- 0x458, 'j',
- 0x43a, 'k',
- 0x459, 'l'+('j'<<16),
- 0x43b, 'l',
- 0x43c, 'm',
- 0x45a, 'n'+('j'<<16),
- 0x43d, 'n',
- 0x43e, 'o',
- 0x43f, 'p',
- 0x440, 'r',
- 0x441, 's',
- 0x448, 0x161,
- 0x442, 't',
- 0x443, 'u',
- 0x432, 'v',
- 0x437, 'z',
- 0x436, 0x17e,
- 0x453, 0x111,
- 0x45c, 0x107,
- 0 }; // ѓ ѕ ќ
- static const unsigned char ru_vowels[] = { 0x10, 0x15, 0x31, 0x18, 0x1e, 0x23, 0x2b, 0x2d, 0x2e, 0x2f, 0xb9, 0xc9, 0x91, 0x8f, 0x36, 0 }; // also kazakh
- static const unsigned char ru_consonants[] = { 0x11, 0x12, 0x13, 0x14, 0x16, 0x17, 0x19, 0x1a, 0x1b, 0x1c, 0x1d, 0x1f, 0x20, 0x21, 0x22, 0x24, 0x25, 0x26, 0x27, 0x28, 0x29, 0x2a, 0x2c, 0x73, 0x7b, 0x83, 0x9b, 0 };
- static void SetCyrillicLetters(Translator *tr)
- {
- // character codes offset by 0x420
- static const char ru_soft[] = { 0x2c, 0x19, 0x27, 0x29, 0 }; // letter group B [k ts; s;]
- static const char ru_hard[] = { 0x2a, 0x16, 0x26, 0x28, 0 }; // letter group H [S Z ts]
- static const char ru_nothard[] = { 0x11, 0x12, 0x13, 0x14, 0x17, 0x19, 0x1a, 0x1b, 0x1c, 0x1d, 0x1f, 0x20, 0x21, 0x22, 0x24, 0x25, 0x27, 0x29, 0x2c, 0 };
- static const char ru_voiced[] = { 0x11, 0x12, 0x13, 0x14, 0x16, 0x17, 0 }; // letter group G (voiced obstruents)
- static const char ru_ivowels[] = { 0x2c, 0x2e, 0x2f, 0x31, 0 }; // letter group Y (iotated vowels & soft-sign)
- tr->encoding = ESPEAKNG_ENCODING_KOI8_R;
- tr->transpose_min = 0x430; // convert cyrillic from unicode into range 0x01 to 0x22
- tr->transpose_max = 0x451;
- tr->transpose_map = NULL;
- tr->frequent_pairs = pairs_ru;
- tr->letter_bits_offset = OFFSET_CYRILLIC;
- memset(tr->letter_bits, 0, sizeof(tr->letter_bits));
- SetLetterBits(tr, LETTERGP_A, (char *)ru_vowels);
- SetLetterBits(tr, LETTERGP_B, ru_soft);
- SetLetterBits(tr, LETTERGP_C, (char *)ru_consonants);
- SetLetterBits(tr, LETTERGP_H, ru_hard);
- SetLetterBits(tr, LETTERGP_F, ru_nothard);
- SetLetterBits(tr, LETTERGP_G, ru_voiced);
- SetLetterBits(tr, LETTERGP_Y, ru_ivowels);
- SetLetterBits(tr, LETTERGP_VOWEL2, (char *)ru_vowels);
- }
- static void SetIndicLetters(Translator *tr)
- {
- // Set letter types for Indic scripts, Devanagari, Tamill, etc
- static const char dev_consonants2[] = { 0x02, 0x03, 0x58, 0x59, 0x5a, 0x5b, 0x5c, 0x5d, 0x5e, 0x5f, 0x7b, 0x7c, 0x7e, 0x7f, 0 };
- static const char dev_vowels2[] = { 0x60, 0x61, 0x55, 0x56, 0x57, 0x62, 0x63, 0 }; // non-consecutive vowels and vowel-signs
- memset(tr->letter_bits, 0, sizeof(tr->letter_bits));
- SetLetterBitsRange(tr, LETTERGP_A, 0x04, 0x14); // vowel letters
- SetLetterBitsRange(tr, LETTERGP_A, 0x3e, 0x4d); // + vowel signs, and virama
- SetLetterBits(tr, LETTERGP_A, dev_vowels2); // + extra vowels and vowel signs
- SetLetterBitsRange(tr, LETTERGP_B, 0x3e, 0x4d); // vowel signs, and virama
- SetLetterBits(tr, LETTERGP_B, dev_vowels2); // + extra vowels and vowel signs
- SetLetterBitsRange(tr, LETTERGP_C, 0x15, 0x39); // the main consonant range
- SetLetterBits(tr, LETTERGP_C, dev_consonants2); // + additional consonants
- SetLetterBitsRange(tr, LETTERGP_Y, 0x04, 0x14); // vowel letters
- SetLetterBitsRange(tr, LETTERGP_Y, 0x3e, 0x4c); // + vowel signs
- SetLetterBits(tr, LETTERGP_Y, dev_vowels2); // + extra vowels and vowel signs
- tr->langopts.param[LOPT_UNPRONOUNCABLE] = 1; // disable check for unpronouncable words
- tr->langopts.suffix_add_e = tr->letter_bits_offset + 0x4d; // virama
- }
- static void SetupTranslator(Translator *tr, const short *lengths, const unsigned char *amps)
- {
- if (lengths != NULL)
- memcpy(tr->stress_lengths, lengths, sizeof(tr->stress_lengths));
- if (amps != NULL)
- memcpy(tr->stress_amps, amps, sizeof(tr->stress_amps));
- }
- Translator *SelectTranslator(const char *name)
- {
- int name2 = 0;
- Translator *tr;
- static const short stress_lengths_equal[8] = { 230, 230, 230, 230, 0, 0, 230, 230 };
- static const unsigned char stress_amps_equal[8] = { 19, 19, 19, 19, 19, 19, 19, 19 };
- static const short stress_lengths_fr[8] = { 190, 170, 190, 200, 0, 0, 190, 240 };
- static const unsigned char stress_amps_fr[8] = { 18, 16, 18, 18, 18, 18, 18, 18 };
- static const unsigned char stress_amps_sk[8] = { 17, 16, 20, 20, 20, 22, 22, 21 };
- static const short stress_lengths_sk[8] = { 190, 190, 210, 210, 0, 0, 210, 210 };
- static const short stress_lengths_ta[8] = { 200, 200, 210, 210, 0, 0, 230, 230 };
- static const short stress_lengths_ta2[8] = { 230, 230, 240, 240, 0, 0, 260, 260 };
- static const unsigned char stress_amps_ta[8] = { 18, 18, 18, 18, 20, 20, 22, 22 };
- tr = NewTranslator();
- strcpy(tr->dictionary_name, name);
- // convert name string into a word of up to 4 characters, for the switch()
- while (*name != 0)
- name2 = (name2 << 8) + *name++;
- switch (name2)
- {
- case L('a', 'f'):
- {
- static const short stress_lengths_af[8] = { 170, 140, 220, 220, 0, 0, 250, 270 };
- SetupTranslator(tr, stress_lengths_af, NULL);
- tr->langopts.stress_rule = STRESSPOSN_1L;
- tr->langopts.vowel_pause = 0x30;
- tr->langopts.param[LOPT_DIERESES] = 1;
- tr->langopts.param[LOPT_PREFIXES] = 1;
- tr->langopts.numbers = NUM_SWAP_TENS | NUM_HUNDRED_AND | NUM_SINGLE_AND | NUM_ROMAN | NUM_1900;
- tr->langopts.accents = 1;
- }
- break;
- case L('a', 'm'): // Amharic, Ethiopia
- {
- SetupTranslator(tr, stress_lengths_fr, stress_amps_fr);
- tr->letter_bits_offset = OFFSET_ETHIOPIC;
- tr->langopts.stress_rule = STRESSPOSN_1L;
- tr->langopts.stress_flags = S_NO_AUTO_2 | S_FINAL_DIM; // don't use secondary stress
- tr->langopts.length_mods0 = tr->langopts.length_mods; // don't lengthen vowels in the last syllable
- tr->langopts.param[LOPT_UNPRONOUNCABLE] = 1; // disable check for unpronouncable words
- tr->langopts.numbers = NUM_OMIT_1_HUNDRED;
- }
- break;
- case L('a', 'r'): // Arabic
- tr->transpose_min = 0x620; // for ar_list, use 6-bit character codes
- tr->transpose_max = 0x65f;
- tr->transpose_map = NULL;
- tr->letter_bits_offset = OFFSET_ARABIC;
- tr->langopts.numbers = NUM_SWAP_TENS | NUM_AND_UNITS | NUM_HUNDRED_AND | NUM_OMIT_1_HUNDRED | NUM_AND_HUNDRED | NUM_THOUSAND_AND | NUM_OMIT_1_THOUSAND;
- tr->langopts.param[LOPT_UNPRONOUNCABLE] = 1; // disable check for unpronouncable words
- tr->encoding = ESPEAKNG_ENCODING_ISO_8859_6;
- break;
- case L('b', 'g'): // Bulgarian
- {
- SetCyrillicLetters(tr);
- tr->encoding = ESPEAKNG_ENCODING_ISO_8859_5;
- tr->langopts.param[LOPT_UNPRONOUNCABLE] = 0x432; // [v] don't count this character at start of word
- tr->langopts.param[LOPT_REGRESSIVE_VOICING] = 0x107; // devoice at end of word, and change voicing to match a following consonant (except v)
- tr->langopts.param[LOPT_REDUCE] = 2;
- tr->langopts.stress_rule = STRESSPOSN_2R;
- tr->langopts.numbers = NUM_DECIMAL_COMMA | NUM_ALLOW_SPACE | NUM_OMIT_1_HUNDRED | NUM_HUNDRED_AND | NUM_AND_UNITS | NUM_SINGLE_AND | NUM_ROMAN | NUM_ROMAN_ORDINAL | NUM_ROMAN_CAPITALS;
- tr->langopts.thousands_sep = ' '; // don't allow dot as thousands separator
- }
- break;
- case L('b', 'n'): // Bengali
- case L('a', 's'): // Assamese
- case L3('b', 'p', 'y'): // Manipuri (temporary placement - it's not indo-european)
- {
- static const short stress_lengths_bn[8] = { 180, 180, 210, 210, 0, 0, 230, 240 };
- static const unsigned char stress_amps_bn[8] = { 18, 18, 18, 18, 20, 20, 22, 22 };
- static const char bn_consonants2[3] = { 0x70, 0x71, 0 };
- SetupTranslator(tr, stress_lengths_bn, stress_amps_bn);
- tr->langopts.length_mods0 = tr->langopts.length_mods; // don't lengthen vowels in the last syllable
- tr->langopts.stress_rule = STRESSPOSN_1L;
- tr->langopts.stress_flags = S_MID_DIM | S_FINAL_DIM; // use 'diminished' for unstressed final syllable
- tr->letter_bits_offset = OFFSET_BENGALI;
- SetIndicLetters(tr); // call this after setting OFFSET_BENGALI
- SetLetterBitsRange(tr, LETTERGP_B, 0x01, 0x01); // candranindu
- SetLetterBitsRange(tr, LETTERGP_F, 0x3e, 0x4c); // vowel signs, but not virama
- SetLetterBits(tr, LETTERGP_C, bn_consonants2);
- tr->langopts.numbers = NUM_SWAP_TENS;
- tr->langopts.break_numbers = 0x24924aa8; // for languages which have numbers for 100,000 and 100,00,000, eg Hindi
- if (name2 == L3('b', 'p', 'y')) {
- tr->langopts.numbers = 1;
- tr->langopts.numbers2 = NUM2_SWAP_THOUSANDS;
- }
- }
- break;
- case L('b', 'o'): // Tibet
- {
- tr->langopts.stress_rule = STRESSPOSN_1L;
- tr->letter_bits_offset = OFFSET_TIBET;
- SetLetterBitsRange(tr, LETTERGP_A, 0x71, 0x7d); // vowel signs
- SetLetterBitsRange(tr, LETTERGP_B, 0x71, 0x81); // vowel signs and subjoined letters
- SetLetterBitsRange(tr, LETTERGP_B, 0x90, 0xbc);
- SetLetterBitsRange(tr, LETTERGP_C, 0x40, 0x6c); // consonant letters (not subjoined)
- tr->langopts.param[LOPT_UNPRONOUNCABLE] = 1; // disable check for unpronouncable words
- tr->langopts.numbers = 1;
- }
- break;
- case L('c', 'y'): // Welsh
- {
- static const short stress_lengths_cy[8] = { 170, 220, 180, 180, 0, 0, 250, 270 };
- static const unsigned char stress_amps_cy[8] = { 17, 15, 18, 18, 0, 0, 22, 20 }; // 'diminished' is used to mark a quieter, final unstressed syllable
- SetupTranslator(tr, stress_lengths_cy, stress_amps_cy);
- tr->encoding = ESPEAKNG_ENCODING_ISO_8859_14;
- tr->langopts.stress_rule = STRESSPOSN_2R;
- // 'diminished' is an unstressed final syllable
- tr->langopts.stress_flags = S_FINAL_DIM_ONLY | S_FINAL_NO_2;
- tr->langopts.unstressed_wd1 = 0;
- tr->langopts.unstressed_wd2 = 2;
- tr->langopts.param[LOPT_SONORANT_MIN] = 120; // limit the shortening of sonorants before short vowels
- tr->langopts.numbers = NUM_OMIT_1_HUNDRED;
- }
- break;
- case L('d', 'a'): // Danish
- {
- static const short stress_lengths_da[8] = { 160, 140, 200, 200, 0, 0, 220, 230 };
- SetupTranslator(tr, stress_lengths_da, NULL);
- tr->langopts.stress_rule = STRESSPOSN_1L;
- tr->langopts.param[LOPT_PREFIXES] = 1;
- tr->langopts.numbers = NUM_DECIMAL_COMMA | NUM_SWAP_TENS | NUM_HUNDRED_AND | NUM_OMIT_1_HUNDRED | NUM_ORDINAL_DOT | NUM_1900 | NUM_ROMAN | NUM_ROMAN_CAPITALS | NUM_ROMAN_ORDINAL;
- }
- break;
- case L('d', 'e'):
- {
- static const short stress_lengths_de[8] = { 150, 130, 200, 200, 0, 0, 270, 270 };
- static const unsigned char stress_amps_de[] = { 20, 20, 20, 20, 20, 22, 22, 20 };
- SetupTranslator(tr, stress_lengths_de, stress_amps_de);
- tr->langopts.stress_rule = STRESSPOSN_1L;
- tr->langopts.word_gap = 0x8; // don't use linking phonemes
- tr->langopts.vowel_pause = 0x30;
- tr->langopts.param[LOPT_PREFIXES] = 1;
- tr->langopts.param[LOPT_REGRESSIVE_VOICING] = 0x100; // devoice at end of word
- tr->langopts.param[LOPT_LONG_VOWEL_THRESHOLD] = 175/2;
- tr->langopts.numbers = NUM_DECIMAL_COMMA | NUM_SWAP_TENS | NUM_ALLOW_SPACE | NUM_ORDINAL_DOT | NUM_ROMAN;
- tr->langopts.param[LOPT_UNPRONOUNCABLE] = 2; // use de_rules for unpronouncable rules
- }
- break;
- case L('d', 'v'): // Divehi (Maldives) FIXME: this language code is actually never used
- {
- SetupTranslator(tr, stress_lengths_ta, stress_amps_ta);
- tr->langopts.param[LOPT_UNPRONOUNCABLE] = 1; // disable check for unpronouncable words
- tr->langopts.length_mods0 = tr->langopts.length_mods; // don't lengthen vowels in the last syllable
- tr->letter_bits_offset = OFFSET_THAANA;
- tr->langopts.stress_rule = STRESSPOSN_1L;
- tr->langopts.stress_flags = S_MID_DIM | S_FINAL_DIM; // use 'diminished' for unstressed final syllable
- SetLetterBitsRange(tr, LETTERGP_B, 0x26, 0x30); // vowel signs, and virama
- tr->langopts.break_numbers = 0x14a8; // 1000, 100,000 10,000,000
- tr->langopts.numbers = 1;
- }
- break;
- case L('e', 'n'):
- {
- static const short stress_lengths_en[8] = { 182, 140, 220, 220, 0, 0, 248, 275 };
- SetupTranslator(tr, stress_lengths_en, NULL);
- tr->langopts.stress_rule = STRESSPOSN_1L;
- tr->langopts.stress_flags = 0x08;
- tr->langopts.numbers = NUM_HUNDRED_AND | NUM_ROMAN | NUM_1900;
- tr->langopts.param[LOPT_COMBINE_WORDS] = 2; // allow "mc" to cmbine with the following word
- tr->langopts.suffix_add_e = 'e';
- tr->langopts.param[LOPT_UNPRONOUNCABLE] = 2; // use en_rules for unpronouncable rules
- SetLetterBits(tr, 6, "aeiouy"); // Group Y: vowels, including y
- }
- break;
- case L('e', 'l'): // Greek
- case L3('g', 'r', 'c'): // Ancient Greek
- {
- static const short stress_lengths_el[8] = { 155, 180, 210, 210, 0, 0, 270, 300 };
- static const unsigned char stress_amps_el[8] = { 15, 12, 20, 20, 20, 22, 22, 21 }; // 'diminished' is used to mark a quieter, final unstressed syllable
- // character codes offset by 0x380
- static const char el_vowels[] = { 0x10, 0x2c, 0x2d, 0x2e, 0x2f, 0x30, 0x31, 0x35, 0x37, 0x39, 0x3f, 0x45, 0x49, 0x4a, 0x4b, 0x4c, 0x4d, 0x4e, 0x4f, 0 };
- static const char el_fvowels[] = { 0x2d, 0x2e, 0x2f, 0x35, 0x37, 0x39, 0x45, 0x4d, 0 }; // ε η ι υ έ ή ί ύ _
- static const char el_voiceless[] = { 0x38, 0x3a, 0x3e, 0x40, 0x42, 0x43, 0x44, 0x46, 0x47, 0 }; // θ κ ξ π ς σ τ φ χ _
- static const char el_consonants[] = { 0x32, 0x33, 0x34, 0x36, 0x38, 0x3a, 0x3b, 0x3c, 0x3d, 0x3e, 0x40, 0x41, 0x42, 0x43, 0x44, 0x46, 0x47, 0x48, 0 };
- static const wchar_t el_char_apostrophe[] = { 0x3c3, 0 }; // σ _
- SetupTranslator(tr, stress_lengths_el, stress_amps_el);
- tr->encoding = ESPEAKNG_ENCODING_ISO_8859_7;
- tr->char_plus_apostrophe = el_char_apostrophe;
- tr->letter_bits_offset = OFFSET_GREEK;
- memset(tr->letter_bits, 0, sizeof(tr->letter_bits));
- SetLetterBits(tr, LETTERGP_A, el_vowels);
- SetLetterBits(tr, LETTERGP_VOWEL2, el_vowels);
- SetLetterBits(tr, LETTERGP_B, el_voiceless);
- SetLetterBits(tr, LETTERGP_C, el_consonants);
- SetLetterBits(tr, LETTERGP_Y, el_fvowels); // front vowels: ε η ι υ _
- tr->langopts.length_mods0 = tr->langopts.length_mods; // don't lengthen vowels in the last syllable
- tr->langopts.stress_rule = STRESSPOSN_2R;
- tr->langopts.stress_flags = S_FINAL_DIM_ONLY; // mark unstressed final syllables as diminished
- tr->langopts.unstressed_wd1 = 0;
- tr->langopts.unstressed_wd2 = 2;
- tr->langopts.param[LOPT_SONORANT_MIN] = 130; // limit the shortening of sonorants before short vowels
- tr->langopts.numbers = NUM_SINGLE_STRESS | NUM_DECIMAL_COMMA;
- tr->langopts.numbers2 = 0x2 | NUM2_MULTIPLE_ORDINAL | NUM2_ORDINAL_NO_AND; // variant form of numbers before thousands
- if (name2 == L3('g', 'r', 'c')) {
- // ancient greek
- tr->langopts.param[LOPT_UNPRONOUNCABLE] = 1;
- }
- }
- break;
- case L('e', 'o'):
- {
- static const short stress_lengths_eo[8] = { 150, 140, 180, 180, 0, 0, 200, 200 };
- static const unsigned char stress_amps_eo[] = { 16, 14, 20, 20, 20, 22, 22, 21 };
- static const wchar_t eo_char_apostrophe[2] = { 'l', 0 };
- SetupTranslator(tr, stress_lengths_eo, stress_amps_eo);
- tr->encoding = ESPEAKNG_ENCODING_ISO_8859_3;
- tr->char_plus_apostrophe = eo_char_apostrophe;
- tr->langopts.vowel_pause = 2;
- tr->langopts.stress_rule = STRESSPOSN_2R;
- tr->langopts.stress_flags = S_FINAL_DIM_ONLY | S_FINAL_NO_2;
- tr->langopts.unstressed_wd2 = 2;
- tr->langopts.numbers = NUM_DECIMAL_COMMA | NUM_OMIT_1_HUNDRED | NUM_ALLOW_SPACE | NUM_ROMAN;
- }
- break;
- case L('e', 's'): // Spanish
- case L('a', 'n'): // Aragonese
- case L('c', 'a'): // Catalan
- case L('i', 'a'): // Interlingua
- case L3('p', 'a', 'p'): // Papiamento
- {
- static const short stress_lengths_es[8] = { 160, 145, 155, 150, 0, 0, 200, 245 };
- static const unsigned char stress_amps_es[8] = { 16, 14, 15, 16, 20, 20, 22, 22 }; // 'diminished' is used to mark a quieter, final unstressed syllable
- static const wchar_t ca_punct_within_word[] = { '\'', 0xb7, 0 }; // ca: allow middle-dot within word
- SetupTranslator(tr, stress_lengths_es, stress_amps_es);
- tr->langopts.length_mods0 = tr->langopts.length_mods; // don't lengthen vowels in the last syllable
- tr->langopts.stress_rule = STRESSPOSN_2R;
- // stress last syllable if it doesn't end in vowel or "s" or "n"
- // 'diminished' is an unstressed final syllable
- tr->langopts.stress_flags = S_FINAL_SPANISH | S_FINAL_DIM_ONLY | S_FINAL_NO_2;
- tr->langopts.unstressed_wd1 = 0;
- tr->langopts.unstressed_wd2 = 2;
- tr->langopts.param[LOPT_SONORANT_MIN] = 120; // limit the shortening of sonorants before short vowels
- tr->langopts.numbers = NUM_SINGLE_STRESS | NUM_DECIMAL_COMMA | NUM_AND_UNITS | NUM_OMIT_1_HUNDRED | NUM_OMIT_1_THOUSAND | NUM_ROMAN | NUM_ROMAN_AFTER | NUM_DFRACTION_4;
- tr->langopts.numbers2 = NUM2_MULTIPLE_ORDINAL | NUM2_ORDINAL_NO_AND;
- if (name2 == L('c', 'a')) {
- // stress last syllable unless word ends with a vowel
- tr->punct_within_word = ca_punct_within_word;
- tr->langopts.stress_flags = S_FINAL_SPANISH | S_FINAL_DIM_ONLY | S_FINAL_NO_2 | S_NO_AUTO_2;
- } else if (name2 == L('i', 'a')) {
- tr->langopts.stress_flags = S_FINAL_SPANISH | S_FINAL_DIM_ONLY | S_FINAL_NO_2;
- tr->langopts.numbers = NUM_DECIMAL_COMMA | NUM_OMIT_1_HUNDRED | NUM_OMIT_1_THOUSAND | NUM_ROMAN | NUM_ROMAN_AFTER;
- } else if (name2 == L('a', 'n')) {
- tr->langopts.stress_flags = S_FINAL_SPANISH | S_FINAL_DIM_ONLY | S_FINAL_NO_2;
- tr->langopts.numbers = NUM_SINGLE_STRESS | NUM_DECIMAL_COMMA | NUM_AND_UNITS | NUM_OMIT_1_HUNDRED | NUM_OMIT_1_THOUSAND | NUM_ROMAN | NUM_ROMAN_ORDINAL;
- tr->langopts.numbers2 = NUM2_ORDINAL_NO_AND;
- tr->langopts.roman_suffix = utf8_ordinal;
- } else if (name2 == L3('p', 'a', 'p')) {
- // stress last syllable unless word ends with a vowel
- tr->langopts.stress_rule = STRESSPOSN_1R;
- tr->langopts.stress_flags = S_FINAL_VOWEL_UNSTRESSED | S_FINAL_DIM_ONLY | S_FINAL_NO_2 | S_NO_AUTO_2;
- } else
- tr->langopts.param[LOPT_UNPRONOUNCABLE] = 2; // use es_rules for unpronouncable rules
- }
- break;
- case L('e', 'u'): // basque
- {
- static const short stress_lengths_eu[8] = { 200, 200, 200, 200, 0, 0, 210, 230 }; // very weak stress
- static const unsigned char stress_amps_eu[8] = { 16, 16, 18, 18, 18, 18, 18, 18 };
- SetupTranslator(tr, stress_lengths_eu, stress_amps_eu);
- tr->langopts.stress_rule = STRESSPOSN_2L; // ?? second syllable, but not on a word-final vowel
- tr->langopts.stress_flags = S_FINAL_VOWEL_UNSTRESSED;
- tr->langopts.param[LOPT_SUFFIX] = 1;
- tr->langopts.numbers = NUM_SINGLE_STRESS | NUM_DECIMAL_COMMA | NUM_HUNDRED_AND | NUM_OMIT_1_HUNDRED | NUM_OMIT_1_THOUSAND | NUM_VIGESIMAL;
- }
- break;
- case L('f', 'a'): // Farsi
- {
- // Convert characters in the range 0x620 to 0x6cc to the range 1 to 63.
- // 0 indicates no translation for this character
- static const char transpose_map_fa[] = {
- 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, // 0x620
- 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 0, 0, 0, 0, 0, // 0x630
- 0, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, // 0x640
- 42, 43, 0, 0, 44, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, // 0x650
- 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, // 0x660
- 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 45, 0, // 0x670
- 0, 0, 0, 0, 0, 0, 46, 0, 0, 0, 0, 0, 0, 0, 0, 0, // 0x680
- 0, 0, 0, 0, 0, 0, 0, 0, 47, 0, 0, 0, 0, 0, 0, 0, // 0x690
- 0, 0, 0, 0, 0, 0, 0, 0, 0, 48, 0, 0, 0, 0, 0, 49, // 0x6a0
- 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, // 0x6b0
- 50, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 51 // 0x6c0
- };
- tr->transpose_min = 0x620;
- tr->transpose_max = 0x6cc;
- tr->transpose_map = transpose_map_fa;
- tr->letter_bits_offset = OFFSET_ARABIC;
- tr->langopts.numbers = NUM_AND_UNITS | NUM_HUNDRED_AND;
- tr->langopts.param[LOPT_UNPRONOUNCABLE] = 1; // disable check for unpronouncable words
- tr->chars_ignore = chars_ignore_zwnj_hyphen; // replace ZWNJ by hyphen
- }
- break;
- case L('e', 't'): // Estonian
- tr->encoding = ESPEAKNG_ENCODING_ISO_8859_4;
- // fallthrough:
- case L('f', 'i'): // Finnish
- {
- static const unsigned char stress_amps_fi[8] = { 18, 16, 22, 22, 20, 22, 22, 22 };
- static const short stress_lengths_fi[8] = { 150, 180, 200, 200, 0, 0, 210, 250 };
- SetupTranslator(tr, stress_lengths_fi, stress_amps_fi);
- tr->langopts.stress_rule = STRESSPOSN_1L;
- tr->langopts.stress_flags = S_FINAL_DIM_ONLY | S_FINAL_NO_2 | S_2_TO_HEAVY; // move secondary stress from light to a following heavy syllable
- tr->langopts.param[LOPT_IT_DOUBLING] = 1;
- tr->langopts.long_stop = 130;
- tr->langopts.numbers = NUM_DECIMAL_COMMA + NUM_ALLOW_SPACE;
- tr->langopts.spelling_stress = 1;
- tr->langopts.intonation_group = 3; // less intonation, don't raise pitch at comma
- }
- break;
- case L('f', 'r'): // french
- {
- SetupTranslator(tr, stress_lengths_fr, stress_amps_fr);
- tr->langopts.stress_rule = STRESSPOSN_1R; // stress on final syllable
- tr->langopts.stress_flags = S_NO_AUTO_2 | S_FINAL_DIM; // don't use secondary stress
- tr->langopts.param[LOPT_IT_LENGTHEN] = 1; // remove lengthen indicator from unstressed syllables
- tr->langopts.length_mods0 = tr->langopts.length_mods; // don't lengthen vowels in the last syllable
- tr->langopts.accents = 2; // Say "Capital" after the letter.
- tr->langopts.numbers = NUM_SINGLE_STRESS | NUM_DECIMAL_COMMA | NUM_ALLOW_SPACE | NUM_OMIT_1_HUNDRED | NUM_NOPAUSE | NUM_ROMAN | NUM_ROMAN_CAPITALS | NUM_ROMAN_AFTER | NUM_VIGESIMAL | NUM_DFRACTION_4;
- }
- break;
- case L3('h','a', 'k'): // Hakka Chinese
- {
- tr->langopts.stress_flags = S_NO_DIM; // don't automatically set diminished stress (may be set in the intonation module)
- tr->langopts.tone_language = 1; // Tone language, use CalcPitches_Tone() rather than CalcPitches()
- tr->langopts.tone_numbers = 1; // a number after letters indicates a tone number (eg. pinyin or jyutping)
- tr->langopts.ideographs = 1;
- }
- break;
- case L('g', 'a'): // irish
- case L('g', 'd'): // scots gaelic
- {
- tr->langopts.stress_rule = STRESSPOSN_1L;
- tr->langopts.stress_flags = S_NO_AUTO_2; // don't use secondary stress
- tr->langopts.numbers = NUM_OMIT_1_HUNDRED | NUM_OMIT_1_THOUSAND;
- tr->langopts.accents = 2; // 'capital' after letter name
- tr->langopts.param[LOPT_UNPRONOUNCABLE] = 3; // don't count apostrophe
- tr->langopts.param[LOPT_IT_LENGTHEN] = 1; // remove [:] phoneme from non-stressed syllables (Lang=gd)
- }
- break;
- case L('g','n'): // guarani
- {
- tr->langopts.stress_rule = STRESSPOSN_1R; // stress on final syllable
- tr->langopts.length_mods0 = tr->langopts.length_mods; // don't lengthen vowels in the last syllable
- }
- break;
- case L('h', 'i'): // Hindi
- case L('n', 'e'): // Nepali
- case L('o', 'r'): // Oriya
- case L('p', 'a'): // Punjabi
- case L('g', 'u'): // Gujarati
- case L('m', 'r'): // Marathi
- {
- static const short stress_lengths_hi[8] = { 190, 190, 210, 210, 0, 0, 230, 250 };
- static const unsigned char stress_amps_hi[8] = { 17, 14, 20, 19, 20, 22, 22, 21 };
- SetupTranslator(tr, stress_lengths_hi, stress_amps_hi);
- tr->encoding = ESPEAKNG_ENCODING_ISCII;
- tr->langopts.length_mods0 = tr->langopts.length_mods; // don't lengthen vowels in the last syllable
- tr->langopts.stress_rule = 6; // stress on last heaviest syllable, excluding final syllable
- tr->langopts.stress_flags = S_MID_DIM | S_FINAL_DIM; // use 'diminished' for unstressed final syllable
- tr->langopts.numbers = NUM_SWAP_TENS;
- tr->langopts.break_numbers = 0x14aa8; // for languages which have numbers for 100,000 and 100,00,000, eg Hindi
- tr->letter_bits_offset = OFFSET_DEVANAGARI;
- if (name2 == L('p', 'a'))
- tr->letter_bits_offset = OFFSET_GURMUKHI;
- else if (name2 == L('g', 'u')) {
- SetupTranslator(tr, stress_lengths_equal, stress_amps_equal);
- tr->letter_bits_offset = OFFSET_GUJARATI;
- tr->langopts.stress_rule = STRESSPOSN_2R;
- } else if (name2 == L('n', 'e')) {
- SetupTranslator(tr, stress_lengths_equal, stress_amps_equal);
- tr->langopts.break_numbers = 0x2aaaa8;
- tr->langopts.max_digits = 22;
- tr->langopts.numbers2 |= NUM2_ENGLISH_NUMERALS;
- } else if (name2 == L('o', 'r'))
- tr->letter_bits_offset = OFFSET_ORIYA;
- SetIndicLetters(tr);
- }
- break;
- case L('h', 'r'): // Croatian
- case L('b', 's'): // Bosnian
- case L('s', 'r'): // Serbian
- {
- static const unsigned char stress_amps_hr[8] = { 17, 17, 20, 20, 20, 22, 22, 21 };
- static const short stress_lengths_hr[8] = { 180, 160, 200, 200, 0, 0, 220, 230 };
- static const short stress_lengths_sr[8] = { 160, 150, 200, 200, 0, 0, 250, 260 };
- strcpy(tr->dictionary_name, "hbs");
- if (name2 == L('s', 'r'))
- SetupTranslator(tr, stress_lengths_sr, stress_amps_hr);
- else
- SetupTranslator(tr, stress_lengths_hr, stress_amps_hr);
- tr->encoding = ESPEAKNG_ENCODING_ISO_8859_2;
- tr->langopts.stress_rule = STRESSPOSN_1L;
- tr->langopts.stress_flags = S_FINAL_NO_2;
- tr->langopts.param[LOPT_REGRESSIVE_VOICING] = 0x3;
- tr->langopts.max_initial_consonants = 5;
- tr->langopts.spelling_stress = 1;
- tr->langopts.accents = 1;
- tr->langopts.numbers = NUM_SINGLE_STRESS | NUM_HUNDRED_AND | NUM_OMIT_1_HUNDRED | NUM_DECIMAL_COMMA | NUM_THOUS_SPACE | NUM_DFRACTION_2 | NUM_ROMAN_CAPITALS;
- tr->langopts.numbers2 = 0xa + NUM2_THOUSANDS_VAR5; // variant numbers before thousands,milliards
- tr->langopts.replace_chars = replace_cyrillic_latin;
- tr->langopts.our_alphabet = OFFSET_CYRILLIC; // don't say "cyrillic" before letter names
- }
- break;
- case L('h', 't'): // Haitian Creole
- tr->langopts.stress_rule = STRESSPOSN_1R; // stress on final syllable
- tr->langopts.stress_flags = S_NO_AUTO_2 | S_FINAL_DIM; // don't use secondary stress
- tr->langopts.numbers = NUM_SINGLE_STRESS | NUM_OMIT_1_HUNDRED | NUM_NOPAUSE | NUM_ROMAN | NUM_VIGESIMAL | NUM_DFRACTION_4;
- break;
- case L('h', 'u'): // Hungarian
- {
- static const unsigned char stress_amps_hu[8] = { 17, 17, 19, 19, 20, 22, 22, 21 };
- static const short stress_lengths_hu[8] = { 185, 195, 195, 190, 0, 0, 210, 220 };
- SetupTranslator(tr, stress_lengths_hu, stress_amps_hu);
- tr->encoding = ESPEAKNG_ENCODING_ISO_8859_2;
- tr->langopts.vowel_pause = 0x20;
- tr->langopts.stress_rule = STRESSPOSN_1L;
- tr->langopts.stress_flags = S_FINAL_DIM_ONLY | S_FINAL_NO_2 | S_NO_AUTO_2 | 0x8000 | S_HYPEN_UNSTRESS;
- tr->langopts.unstressed_wd1 = 2;
- tr->langopts.param[LOPT_IT_DOUBLING] = 1;
- tr->langopts.param[LOPT_ANNOUNCE_PUNCT] = 2; // don't break clause before announcing . ? !
- tr->langopts.numbers = NUM_DFRACTION_5 | NUM_ALLOW_SPACE | NUM_ROMAN | NUM_ROMAN_ORDINAL | NUM_ROMAN_CAPITALS | NUM_ORDINAL_DOT | NUM_OMIT_1_HUNDRED | NUM_OMIT_1_THOUSAND;
- tr->langopts.thousands_sep = ' '; // don't allow dot as thousands separator
- tr->langopts.decimal_sep = ',';
- tr->langopts.max_roman = 899;
- tr->langopts.min_roman = 1;
- tr->langopts.spelling_stress = 1;
- SetLengthMods(tr, 3); // all equal
- }
- break;
- case L('h', 'y'): // Armenian
- {
- static const short stress_lengths_hy[8] = { 250, 200, 250, 250, 0, 0, 250, 250 };
- static const char hy_vowels[] = { 0x31, 0x35, 0x37, 0x38, 0x3b, 0x48, 0x55, 0 };
- static const char hy_consonants[] = {
- 0x32, 0x33, 0x34, 0x36, 0x39, 0x3a, 0x3c, 0x3d, 0x3e, 0x3f, 0x40, 0x41, 0x42, 0x43, 0x44,
- 0x46, 0x47, 0x49, 0x4a, 0x4b, 0x4c, 0x4d, 0x4e, 0x4f, 0x50, 0x51, 0x52, 0x53, 0x54, 0x56, 0
- };
- static const char hy_consonants2[] = { 0x45, 0 };
- SetupTranslator(tr, stress_lengths_hy, NULL);
- tr->langopts.stress_rule = STRESSPOSN_1R; // default stress on final syllable
- tr->letter_bits_offset = OFFSET_ARMENIAN;
- memset(tr->letter_bits, 0, sizeof(tr->letter_bits));
- SetLetterBits(tr, LETTERGP_A, hy_vowels);
- SetLetterBits(tr, LETTERGP_VOWEL2, hy_vowels);
- SetLetterBits(tr, LETTERGP_B, hy_consonants); // not including 'j'
- SetLetterBits(tr, LETTERGP_C, hy_consonants);
- SetLetterBits(tr, LETTERGP_C, hy_consonants2); // add 'j'
- tr->langopts.max_initial_consonants = 6;
- tr->langopts.numbers = NUM_DECIMAL_COMMA | NUM_ALLOW_SPACE | NUM_OMIT_1_HUNDRED;
- }
- break;
- case L('i', 'd'): // Indonesian
- case L('m', 's'): // Malay
- {
- static const short stress_lengths_id[8] = { 160, 200, 180, 180, 0, 0, 220, 240 };
- static const unsigned char stress_amps_id[8] = { 16, 18, 18, 18, 20, 22, 22, 21 };
- SetupTranslator(tr, stress_lengths_id, stress_amps_id);
- tr->langopts.stress_rule = STRESSPOSN_2R;
- tr->langopts.numbers = NUM_DECIMAL_COMMA | NUM_ALLOW_SPACE | NUM_ROMAN;
- tr->langopts.stress_flags = S_FINAL_DIM_ONLY | S_FINAL_NO_2;
- tr->langopts.accents = 2; // "capital" after letter name
- }
- break;
- case L('i', 's'): // Icelandic
- {
- static const short stress_lengths_is[8] = { 180, 160, 200, 200, 0, 0, 240, 250 };
- static const wchar_t is_lettergroup_B[] = { 'c', 'f', 'h', 'k', 'p', 't', 'x', 0xfe, 0 }; // voiceless conants, including 'þ' ?? 's'
- SetupTranslator(tr, stress_lengths_is, NULL);
- tr->langopts.stress_rule = STRESSPOSN_1L;
- tr->langopts.stress_flags = S_FINAL_NO_2;
- tr->langopts.param[LOPT_IT_LENGTHEN] = 0x11; // remove lengthen indicator from unstressed vowels
- tr->langopts.param[LOPT_REDUCE] = 2;
- ResetLetterBits(tr, 0x18);
- SetLetterBits(tr, 4, "kpst"); // Letter group F
- SetLetterBits(tr, 3, "jvr"); // Letter group H
- tr->letter_groups[1] = is_lettergroup_B;
- tr->langopts.numbers = NUM_DECIMAL_COMMA | NUM_SINGLE_AND | NUM_HUNDRED_AND | NUM_AND_UNITS | NUM_1900;
- tr->langopts.numbers2 = 0x2;
- }
- break;
- case L('i', 't'): // Italian
- {
- static const short stress_lengths_it[8] = { 160, 140, 150, 165, 0, 0, 218, 305 };
- static const unsigned char stress_amps_it[8] = { 17, 15, 18, 16, 20, 22, 22, 22 };
- SetupTranslator(tr, stress_lengths_it, stress_amps_it);
- tr->langopts.length_mods0 = tr->langopts.length_mods; // don't lengthen vowels in the last syllable
- tr->langopts.stress_rule = STRESSPOSN_2R;
- tr->langopts.stress_flags = S_NO_AUTO_2 | S_FINAL_DIM_ONLY | S_PRIORITY_STRESS;
- tr->langopts.vowel_pause = 1;
- tr->langopts.unstressed_wd1 = 0;
- tr->langopts.unstressed_wd2 = 2;
- tr->langopts.param[LOPT_IT_LENGTHEN] = 2; // remove lengthen indicator from unstressed or non-penultimate syllables
- tr->langopts.param[LOPT_IT_DOUBLING] = 1; // double the first consonant if the previous word ends in a stressed vowel (changed to =1, 23.01.2014 - only use if prev.word has $double)
- tr->langopts.param[LOPT_SONORANT_MIN] = 130; // limit the shortening of sonorants before short vowels
- tr->langopts.param[LOPT_REDUCE] = 1; // reduce vowels even if phonemes are specified in it_list
- tr->langopts.param[LOPT_ALT] = 2; // call ApplySpecialAttributes2() if a word has $alt or $alt2
- tr->langopts.numbers = NUM_SINGLE_VOWEL | NUM_OMIT_1_HUNDRED |NUM_DECIMAL_COMMA | NUM_DFRACTION_1 | NUM_ROMAN | NUM_ROMAN_CAPITALS | NUM_ROMAN_ORDINAL;
- tr->langopts.numbers2 = NUM2_NO_TEEN_ORDINALS;
- tr->langopts.roman_suffix = utf8_ordinal;
- tr->langopts.accents = 2; // Say "Capital" after the letter.
- }
- break;
- case L3('j', 'b', 'o'): // Lojban
- {
- static const short stress_lengths_jbo[8] = { 145, 145, 170, 160, 0, 0, 330, 350 };
- static const wchar_t jbo_punct_within_word[] = { '.', ',', '\'', 0x2c8, 0 }; // allow period and comma within a word, also stress marker (from LOPT_CAPS_IN_WORD)
- SetupTranslator(tr, stress_lengths_jbo, NULL);
- tr->langopts.stress_rule = STRESSPOSN_2R;
- tr->langopts.vowel_pause = 0x20c; // pause before a word which starts with a vowel, or after a word which ends in a consonant
- tr->punct_within_word = jbo_punct_within_word;
- tr->langopts.param[LOPT_CAPS_IN_WORD] = 2; // capitals indicate stressed syllables
- tr->langopts.max_lengthmod = 368;
- }
- break;
- case L('k', 'a'): // Georgian
- {
- // character codes offset by 0x1080
- static const char ka_vowels[] = { 0x30, 0x34, 0x38, 0x3d, 0x43, 0x55, 0x57, 0 };
- static const char ka_consonants[] =
- { 0x31, 0x32, 0x33, 0x35, 0x36, 0x37, 0x39, 0x3a, 0x3b, 0x3c, 0x3e, 0x3f, 0x40, 0x41, 0x42, 0x44,
- 0x45, 0x46, 0x47, 0x48, 0x49, 0x4a, 0x4b, 0x4c, 0x4d, 0x4e, 0x4f, 0x50, 0x51, 0x52, 0x53, 0x54, 0x56, 0 };
- SetupTranslator(tr, stress_lengths_ta, stress_amps_ta);
- memset(tr->letter_bits, 0, sizeof(tr->letter_bits));
- SetLetterBits(tr, LETTERGP_A, ka_vowels);
- SetLetterBits(tr, LETTERGP_C, ka_consonants);
- SetLetterBits(tr, LETTERGP_VOWEL2, ka_vowels);
- tr->langopts.stress_rule = STRESSPOSN_1L;
- tr->langopts.stress_flags = S_FINAL_NO_2;
- tr->letter_bits_offset = OFFSET_GEORGIAN;
- tr->langopts.max_initial_consonants = 7;
- tr->langopts.numbers = NUM_VIGESIMAL | NUM_AND_UNITS | NUM_OMIT_1_HUNDRED |NUM_OMIT_1_THOUSAND | NUM_DFRACTION_5 | NUM_ROMAN;
- tr->langopts.alt_alphabet = OFFSET_CYRILLIC;
- tr->langopts.alt_alphabet_lang = L('r', 'u');
- }
- break;
- case L('k', 'k'): // Kazakh
- {
- static const unsigned char stress_amps_tr[8] = { 18, 16, 20, 21, 20, 21, 21, 20 };
- static const short stress_lengths_tr[8] = { 190, 180, 230, 230, 0, 0, 250, 250 };
- tr->letter_bits_offset = OFFSET_CYRILLIC;
- memset(tr->letter_bits, 0, sizeof(tr->letter_bits));
- SetLetterBits(tr, LETTERGP_A, (char *)ru_vowels);
- SetLetterBits(tr, LETTERGP_C, (char *)ru_consonants);
- SetLetterBits(tr, LETTERGP_VOWEL2, (char *)ru_vowels);
- SetupTranslator(tr, stress_lengths_tr, stress_amps_tr);
- tr->langopts.stress_rule = 7; // stress on the last syllable, before any explicitly unstressed syllable
- tr->langopts.stress_flags = S_NO_AUTO_2 + S_NO_EOC_LENGTHEN; // no automatic secondary stress, don't lengthen at end-of-clause
- tr->langopts.lengthen_tonic = 0;
- tr->langopts.param[LOPT_SUFFIX] = 1;
- tr->langopts.numbers = NUM_OMIT_1_HUNDRED | NUM_DFRACTION_6;
- tr->langopts.max_initial_consonants = 2;
- SetLengthMods(tr, 3); // all equal
- }
- break;
- case L('k', 'l'): // Greenlandic
- {
- SetupTranslator(tr, stress_lengths_equal, stress_amps_equal);
- tr->langopts.stress_rule = 12;
- tr->langopts.stress_flags = S_NO_AUTO_2;
- tr->langopts.numbers = NUM_DECIMAL_COMMA | NUM_SWAP_TENS | NUM_HUNDRED_AND | NUM_OMIT_1_HUNDRED | NUM_ORDINAL_DOT | NUM_1900 | NUM_ROMAN | NUM_ROMAN_CAPITALS | NUM_ROMAN_ORDINAL;
- }
- break;
- case L('k', 'o'): // Korean, TEST
- {
- static const char ko_ivowels[] = { 0x63, 0x64, 0x67, 0x68, 0x6d, 0x72, 0x74, 0x75, 0 }; // y and i vowels
- static const unsigned char ko_voiced[] = { 0x02, 0x05, 0x06, 0xab, 0xaf, 0xb7, 0xbc, 0 }; // voiced consonants, l,m,n,N
- tr->letter_bits_offset = OFFSET_KOREAN;
- tr->langopts.our_alphabet = 0xa700;
- memset(tr->letter_bits, 0, sizeof(tr->letter_bits));
- SetLetterBitsRange(tr, LETTERGP_A, 0x61, 0x75);
- SetLetterBits(tr, LETTERGP_Y, ko_ivowels);
- SetLetterBits(tr, LETTERGP_G, (const char *)ko_voiced);
- tr->langopts.stress_rule = 8; // ?? 1st syllable if it is heavy, else 2nd syllable
- tr->langopts.param[LOPT_UNPRONOUNCABLE] = 1; // disable check for unpronouncable words
- tr->langopts.numbers = NUM_OMIT_1_HUNDRED;
- tr->langopts.numbers2 = NUM2_MYRIADS;
- tr->langopts.break_numbers = 0x1111110;
- tr->langopts.max_digits = 20;
- }
- break;
- case L('k', 'u'): // Kurdish
- {
- static const unsigned char stress_amps_ku[8] = { 18, 18, 20, 20, 20, 22, 22, 21 };
- static const short stress_lengths_ku[8] = { 180, 180, 190, 180, 0, 0, 230, 240 };
- SetupTranslator(tr, stress_lengths_ku, stress_amps_ku);
- tr->encoding = ESPEAKNG_ENCODING_ISO_8859_9;
- tr->langopts.stress_rule = 7; // stress on the last syllable, before any explicitly unstressed syllable
- tr->langopts.numbers = NUM_HUNDRED_AND | NUM_AND_UNITS | NUM_OMIT_1_HUNDRED | NUM_AND_HUNDRED;
- tr->langopts.max_initial_consonants = 2;
- }
- break;
- case L('k', 'y'): // Kyrgyx
- tr->langopts.numbers = 1;
- break;
- case L('l', 'a'): // Latin
- {
- tr->encoding = ESPEAKNG_ENCODING_ISO_8859_4; // includes a,e,i,o,u-macron
- tr->langopts.stress_rule = STRESSPOSN_2R;
- tr->langopts.stress_flags = S_NO_AUTO_2;
- tr->langopts.unstressed_wd1 = 0;
- tr->langopts.unstressed_wd2 = 2;
- tr->langopts.param[LOPT_DIERESES] = 1;
- tr->langopts.numbers = NUM_ROMAN;
- tr->langopts.max_roman = 5000;
- }
- break;
- case L('l', 't'): // Lithuanian
- {
- tr->encoding = ESPEAKNG_ENCODING_ISO_8859_4;
- tr->langopts.stress_rule = STRESSPOSN_2R;
- tr->langopts.stress_flags = S_NO_AUTO_2;
- tr->langopts.unstressed_wd1 = 0;
- tr->langopts.unstressed_wd2 = 2;
- tr->langopts.param[LOPT_DIERESES] = 1;
- tr->langopts.numbers = NUM_DECIMAL_COMMA | NUM_OMIT_1_HUNDRED | NUM_DFRACTION_4 | NUM_ORDINAL_DOT;
- tr->langopts.numbers2 = NUM2_THOUSANDS_VAR4;
- tr->langopts.max_roman = 5000;
- }
- break;
- case L('l', 'v'): // latvian
- {
- static const unsigned char stress_amps_lv[8] = { 14, 10, 10, 8, 0, 0, 20, 15 };
- static const short stress_lengths_lv[8] = { 180, 180, 180, 160, 0, 0, 230, 180 };
- SetupTranslator(tr, stress_lengths_lv, stress_amps_lv);
- tr->langopts.stress_rule = STRESSPOSN_1L;
- tr->langopts.spelling_stress = 1;
- tr->encoding = ESPEAKNG_ENCODING_ISO_8859_4;
- tr->langopts.numbers = NUM_DECIMAL_COMMA | NUM_OMIT_1_HUNDRED | NUM_DFRACTION_4 | NUM_ORDINAL_DOT;
- tr->langopts.stress_flags = S_NO_AUTO_2 | S_FINAL_DIM | S_FINAL_DIM_ONLY | S_EO_CLAUSE1;
- }
- break;
- case L('m', 'k'): // Macedonian
- {
- static wchar_t vowels_cyrillic[] = {
- // also include 'р' [R]
- 0x440, 0x430, 0x435, 0x438, 0x439, 0x43e, 0x443, 0x44b, 0x44d,
- 0x44e, 0x44f, 0x450, 0x451, 0x456, 0x457, 0x45d, 0x45e, 0
- };
- static const unsigned char stress_amps_mk[8] = { 17, 17, 20, 20, 20, 22, 22, 21 };
- static const short stress_lengths_mk[8] = { 180, 160, 200, 200, 0, 0, 220, 230 };
- SetupTranslator(tr, stress_lengths_mk, stress_amps_mk);
- tr->encoding = ESPEAKNG_ENCODING_ISO_8859_5;
- tr->letter_groups[0] = tr->letter_groups[7] = vowels_cyrillic;
- tr->letter_bits_offset = OFFSET_CYRILLIC;
- tr->langopts.stress_rule = STRESSPOSN_3R; // antipenultimate
- tr->langopts.numbers = NUM_DECIMAL_COMMA | NUM_AND_UNITS | NUM_OMIT_1_HUNDRED | NUM_OMIT_1_THOUSAND | NUM_DFRACTION_2;
- tr->langopts.numbers2 = 0x8a; // variant numbers before thousands,milliards
- }
- break;
- case L('m', 't'): // Maltese
- {
- tr->encoding = ESPEAKNG_ENCODING_ISO_8859_3;
- tr->langopts.param[LOPT_REGRESSIVE_VOICING] = 0x100; // devoice at end of word
- tr->langopts.stress_rule = STRESSPOSN_2R; // penultimate
- tr->langopts.numbers = 1;
- }
- break;
- case L('n', 'l'): // Dutch
- {
- static const short stress_lengths_nl[8] = { 160, 135, 210, 210, 0, 0, 260, 280 };
- tr->langopts.stress_rule = STRESSPOSN_1L;
- tr->langopts.vowel_pause = 0x30; // ??
- tr->langopts.param[LOPT_DIERESES] = 1;
- tr->langopts.param[LOPT_PREFIXES] = 1;
- tr->langopts.param[LOPT_REGRESSIVE_VOICING] = 0x100; // devoice at end of word
- tr->langopts.numbers = NUM_DECIMAL_COMMA | NUM_SWAP_TENS | NUM_OMIT_1_HUNDRED | NUM_OMIT_1_THOUSAND | NUM_ALLOW_SPACE | NUM_1900 | NUM_ORDINAL_DOT;
- tr->langopts.ordinal_indicator = "e";
- tr->langopts.stress_flags = S_FIRST_PRIMARY;
- memcpy(tr->stress_lengths, stress_lengths_nl, sizeof(tr->stress_lengths));
- }
- break;
- case L('n', 'o'): // Norwegian
- {
- static const short stress_lengths_no[8] = { 160, 140, 200, 200, 0, 0, 220, 230 };
- SetupTranslator(tr, stress_lengths_no, NULL);
- tr->langopts.stress_rule = STRESSPOSN_1L;
- tr->langopts.numbers = NUM_DECIMAL_COMMA | NUM_HUNDRED_AND | NUM_ALLOW_SPACE | NUM_1900 | NUM_ORDINAL_DOT;
- }
- break;
- case L('o', 'm'): // Oromo
- {
- static const unsigned char stress_amps_om[] = { 18, 15, 20, 20, 20, 22, 22, 22 };
- static const short stress_lengths_om[8] = { 200, 200, 200, 200, 0, 0, 200, 200 };
- SetupTranslator(tr, stress_lengths_om, stress_amps_om);
- tr->langopts.stress_rule = STRESSPOSN_2R;
- tr->langopts.stress_flags = S_FINAL_DIM_ONLY | S_FINAL_NO_2 | 0x80000;
- tr->langopts.numbers = NUM_OMIT_1_HUNDRED | NUM_HUNDRED_AND;
- tr->langopts.numbers2 = 0x200; // say "thousands" before its number
- }
- break;
- case L('p', 'l'): // Polish
- {
- static const short stress_lengths_pl[8] = { 160, 190, 175, 175, 0, 0, 200, 210 };
- static const unsigned char stress_amps_pl[8] = { 17, 13, 19, 19, 20, 22, 22, 21 }; // 'diminished' is used to mark a quieter, final unstressed syllable
- SetupTranslator(tr, stress_lengths_pl, stress_amps_pl);
- tr->encoding = ESPEAKNG_ENCODING_ISO_8859_2;
- tr->langopts.stress_rule = STRESSPOSN_2R;
- tr->langopts.stress_flags = S_FINAL_DIM_ONLY; // mark unstressed final syllables as diminished
- tr->langopts.param[LOPT_REGRESSIVE_VOICING] = 0x9;
- tr->langopts.max_initial_consonants = 7; // for example: wchrzczony :)
- tr->langopts.numbers = NUM_DECIMAL_COMMA | NUM_ALLOW_SPACE | NUM_DFRACTION_2;
- tr->langopts.numbers2 = NUM2_THOUSANDS_VAR3;
- tr->langopts.param[LOPT_COMBINE_WORDS] = 4 + 0x100; // combine 'nie' (marked with $alt2) with some 1-syllable (and 2-syllable) words (marked with $alt)
- }
- break;
- case L('p', 't'): // Portuguese
- {
- static const short stress_lengths_pt[8] = { 170, 115, 210, 240, 0, 0, 260, 280 };
- static const unsigned char stress_amps_pt[8] = { 16, 11, 19, 21, 20, 22, 22, 21 }; // 'diminished' is used to mark a quieter, final unstressed syllable
- SetupTranslator(tr, stress_lengths_pt, stress_amps_pt);
- tr->langopts.length_mods0 = tr->langopts.length_mods; // don't lengthen vowels in the last syllable
- tr->langopts.stress_rule = STRESSPOSN_1R; // stress on final syllable
- tr->langopts.stress_flags = S_FINAL_DIM_ONLY | S_FINAL_NO_2 | S_INITIAL_2 | S_PRIORITY_STRESS;
- tr->langopts.numbers = NUM_DECIMAL_COMMA | NUM_DFRACTION_2 | NUM_HUNDRED_AND | NUM_AND_UNITS | NUM_ROMAN_CAPITALS;
- tr->langopts.numbers2 = NUM2_MULTIPLE_ORDINAL | NUM2_NO_TEEN_ORDINALS | NUM2_ORDINAL_NO_AND;
- tr->langopts.max_roman = 5000;
- ResetLetterBits(tr, 0x2);
- SetLetterBits(tr, 1, "bcdfgjkmnpqstvxz"); // B hard consonants, excluding h,l,r,w,y
- tr->langopts.param[LOPT_ALT] = 2; // call ApplySpecialAttributes2() if a word has $alt or $alt2
- tr->langopts.accents = 2; // 'capital' after letter name
- }
- break;
- case L('r', 'o'): // Romanian
- {
- static const short stress_lengths_ro[8] = { 170, 170, 180, 180, 0, 0, 240, 260 };
- static const unsigned char stress_amps_ro[8] = { 15, 13, 18, 18, 20, 22, 22, 21 };
- SetupTranslator(tr, stress_lengths_ro, stress_amps_ro);
- tr->langopts.stress_rule = STRESSPOSN_1R;
- tr->langopts.stress_flags = S_FINAL_VOWEL_UNSTRESSED | S_FINAL_DIM_ONLY;
- tr->encoding = ESPEAKNG_ENCODING_ISO_8859_2;
- tr->langopts.numbers = NUM_DECIMAL_COMMA | NUM_ALLOW_SPACE | NUM_DFRACTION_3 | NUM_AND_UNITS | NUM_ROMAN;
- tr->langopts.numbers2 = 0x1e; // variant numbers before all thousandplex
- }
- break;
- case L('r', 'u'): // Russian
- Translator_Russian(tr);
- break;
- case L('r', 'w'): // Kiryarwanda
- {
- tr->langopts.stress_rule = STRESSPOSN_2R;
- tr->langopts.stress_flags = S_FINAL_DIM_ONLY | S_FINAL_NO_2;
- tr->langopts.length_mods0 = tr->langopts.length_mods; // don't lengthen vowels in the last syllable
- tr->langopts.param[LOPT_UNPRONOUNCABLE] = 1; // disable check for unpronouncable words. Need to allow "bw'" prefix
- tr->langopts.numbers = NUM_HUNDRED_AND | NUM_AND_UNITS | NUM_DFRACTION_2 | NUM_AND_HUNDRED;
- tr->langopts.numbers2 = 0x200; // say "thousands" before its number
- }
- break;
- case L('s', 'k'): // Slovak
- case L('c', 's'): // Czech
- {
- static const char *sk_voiced = "bdgjlmnrvwzaeiouy";
- SetupTranslator(tr, stress_lengths_sk, stress_amps_sk);
- tr->encoding = ESPEAKNG_ENCODING_ISO_8859_2;
- tr->langopts.stress_rule = STRESSPOSN_1L;
- tr->langopts.stress_flags = S_FINAL_DIM_ONLY | S_FINAL_NO_2;
- tr->langopts.param[LOPT_REGRESSIVE_VOICING] = 0x3;
- tr->langopts.max_initial_consonants = 5;
- tr->langopts.spelling_stress = 1;
- tr->langopts.param[LOPT_COMBINE_WORDS] = 4; // combine some prepositions with the following word
- tr->langopts.numbers = NUM_OMIT_1_HUNDRED | NUM_DFRACTION_2 | NUM_ROMAN;
- tr->langopts.numbers2 = NUM2_THOUSANDS_VAR2;
- tr->langopts.thousands_sep = STRESSPOSN_1L; // no thousands separator
- tr->langopts.decimal_sep = ',';
- if (name2 == L('c', 's'))
- tr->langopts.numbers2 = 0x108; // variant numbers before milliards
- ResetLetterBits(tr, 0x20);
- SetLetterBits(tr, 5, sk_voiced);
- }
- break;
- case L('s', 'i'): // Sinhala
- {
- SetupTranslator(tr, stress_lengths_ta, stress_amps_ta);
- tr->langopts.length_mods0 = tr->langopts.length_mods; // don't lengthen vowels in the last syllable
- tr->langopts.stress_rule = STRESSPOSN_1L;
- tr->langopts.stress_flags = S_FINAL_DIM_ONLY | S_FINAL_NO_2;
- tr->langopts.spelling_stress = 1;
- tr->letter_bits_offset = OFFSET_SINHALA;
- memset(tr->letter_bits, 0, sizeof(tr->letter_bits));
- SetLetterBitsRange(tr, LETTERGP_A, 0x05, 0x16); // vowel letters
- SetLetterBitsRange(tr, LETTERGP_A, 0x4a, 0x73); // + vowel signs, and virama
- SetLetterBitsRange(tr, LETTERGP_B, 0x4a, 0x73); // vowel signs, and virama
- SetLetterBitsRange(tr, LETTERGP_C, 0x1a, 0x46); // the main consonant range
- tr->langopts.param[LOPT_UNPRONOUNCABLE] = 1; // disable check for unpronouncable words
- tr->langopts.suffix_add_e = tr->letter_bits_offset + 0x4a; // virama
- tr->langopts.numbers = NUM_OMIT_1_THOUSAND | NUM_SINGLE_STRESS_L | NUM_DFRACTION_7;
- tr->langopts.numbers2 = NUM2_PERCENT_BEFORE;
- tr->langopts.break_numbers = 0x14aa8; // for languages which have numbers for 100,000 and 100,00,000, eg Hindi
- }
- break;
- case L('s', 'l'): // Slovenian
- tr->encoding = ESPEAKNG_ENCODING_ISO_8859_2;
- tr->langopts.stress_rule = STRESSPOSN_2R; // Temporary
- tr->langopts.stress_flags = S_NO_AUTO_2;
- tr->langopts.param[LOPT_REGRESSIVE_VOICING] = 0x103;
- tr->langopts.param[LOPT_UNPRONOUNCABLE] = 0x76; // [v] don't count this character at start of word
- tr->langopts.param[LOPT_ALT] = 2; // call ApplySpecialAttributes2() if a word has $alt or $alt2
- tr->langopts.param[LOPT_IT_LENGTHEN] = 1; // remove lengthen indicator from unstressed syllables
- tr->letter_bits[(int)'r'] |= 0x80; // add 'r' to letter group 7, vowels for Unpronouncable test
- tr->langopts.numbers = NUM_DECIMAL_COMMA | NUM_ALLOW_SPACE | NUM_SWAP_TENS | NUM_OMIT_1_HUNDRED | NUM_DFRACTION_2 | NUM_ORDINAL_DOT | NUM_ROMAN;
- tr->langopts.numbers2 = 0x100; // plural forms of millions etc
- tr->langopts.thousands_sep = ' '; // don't allow dot as thousands separator
- tr->langopts.replace_chars = replace_cyrillic_latin;
- break;
- case L('s', 'q'): // Albanian
- {
- static const short stress_lengths_sq[8] = { 150, 150, 180, 180, 0, 0, 300, 300 };
- static const unsigned char stress_amps_sq[8] = { 16, 12, 16, 16, 20, 20, 21, 19 };
- SetupTranslator(tr, stress_lengths_sq, stress_amps_sq);
- tr->langopts.stress_rule = STRESSPOSN_1R;
- tr->langopts.stress_flags = S_FINAL_DIM_ONLY | S_FINAL_NO_2 | S_FINAL_VOWEL_UNSTRESSED;
- tr->langopts.numbers = NUM_DECIMAL_COMMA | NUM_HUNDRED_AND | NUM_AND_UNITS | NUM_DFRACTION_4;
- tr->langopts.accents = 2; // "capital" after letter name
- }
- break;
- case L('s', 'v'): // Swedish
- {
- static const unsigned char stress_amps_sv[] = { 16, 16, 20, 20, 20, 22, 22, 21 };
- static const short stress_lengths_sv[8] = { 160, 135, 220, 220, 0, 0, 250, 280 };
- SetupTranslator(tr, stress_lengths_sv, stress_amps_sv);
- tr->langopts.stress_rule = STRESSPOSN_1L;
- tr->langopts.numbers = NUM_SINGLE_STRESS | NUM_DECIMAL_COMMA | NUM_ALLOW_SPACE | NUM_1900;
- tr->langopts.accents = 1;
- }
- break;
- case L('s', 'w'): // Swahili
- case L('t', 'n'): // Setswana
- {
- static const short stress_lengths_sw[8] = { 160, 170, 200, 200, 0, 0, 320, 340 };
- static const unsigned char stress_amps_sw[] = { 16, 12, 19, 19, 20, 22, 22, 21 };
- SetupTranslator(tr, stress_lengths_sw, stress_amps_sw);
- tr->langopts.length_mods0 = tr->langopts.length_mods; // don't lengthen vowels in the last syllable
- tr->langopts.vowel_pause = 1;
- tr->langopts.stress_rule = STRESSPOSN_2R;
- tr->langopts.stress_flags = S_FINAL_DIM_ONLY | S_FINAL_NO_2;
- tr->langopts.max_initial_consonants = 4; // for example: mwngi
- tr->langopts.numbers = NUM_AND_UNITS | NUM_HUNDRED_AND | NUM_SINGLE_AND | NUM_OMIT_1_HUNDRED;
- tr->langopts.break_numbers = 0x49249268; // for languages which have numbers for 100,000 and 1,000,000
- }
- break;
- case L('t', 'a'): // Tamil
- case L('k', 'n'): // Kannada
- case L('m', 'l'): // Malayalam
- case L('t', 'e'): // Telugu
- {
- SetupTranslator(tr, stress_lengths_ta2, stress_amps_ta);
- tr->langopts.length_mods0 = tr->langopts.length_mods; // don't lengthen vowels in the last syllable
- tr->langopts.stress_rule = STRESSPOSN_1L;
- tr->langopts.stress_flags = S_FINAL_DIM_ONLY | S_FINAL_NO_2; // use 'diminished' for unstressed final syllable
- tr->langopts.spelling_stress = 1;
- tr->langopts.break_numbers = 0x14a8; // 1000, 100,000 10,000,000
- if (name2 == L('t', 'a')) {
- SetupTranslator(tr, stress_lengths_ta, NULL);
- tr->letter_bits_offset = OFFSET_TAMIL;
- tr->langopts.numbers = NUM_OMIT_1_THOUSAND;
- tr->langopts.numbers2 = NUM2_ORDINAL_AND_THOUSANDS;
- tr->langopts.param[LOPT_WORD_MERGE] = 1; // don't break vowels betwen words
- } else if (name2 == L('m', 'l')) {
- static const short stress_lengths_ml[8] = { 180, 160, 240, 240, 0, 0, 260, 260 };
- SetupTranslator(tr, stress_lengths_ml, stress_amps_equal);
- tr->letter_bits_offset = OFFSET_MALAYALAM;
- tr->langopts.numbers = NUM_OMIT_1_THOUSAND | NUM_OMIT_1_HUNDRED;
- tr->langopts.numbers2 = NUM2_OMIT_1_HUNDRED_ONLY;
- tr->langopts.stress_rule = 13; // 1st syllable, unless 1st vowel is short and 2nd is long
- } else if (name2 == L('k', 'n')) {
- tr->letter_bits_offset = OFFSET_KANNADA;
- tr->langopts.numbers = 0x1;
- } else if (name2 == L('t', 'e')) {
- tr->letter_bits_offset = OFFSET_TELUGU;
- tr->langopts.numbers = 0x1;
- tr->langopts.numbers2 = NUM2_ORDINAL_DROP_VOWEL;
- }
- SetIndicLetters(tr); // call this after setting OFFSET_
- SetLetterBitsRange(tr, LETTERGP_B, 0x4e, 0x4e); // chillu-virama (unofficial)
- }
- break;
- case L('t', 'r'): // Turkish
- case L('a', 'z'): // Azerbaijan
- {
- static const unsigned char stress_amps_tr[8] = { 18, 16, 20, 21, 20, 21, 21, 20 };
- static const short stress_lengths_tr[8] = { 190, 180, 200, 230, 0, 0, 240, 250 };
- SetupTranslator(tr, stress_lengths_tr, stress_amps_tr);
- tr->encoding = ESPEAKNG_ENCODING_ISO_8859_9;
- tr->langopts.stress_rule = 7; // stress on the last syllable, before any explicitly unstressed syllable
- tr->langopts.stress_flags = S_NO_AUTO_2; // no automatic secondary stress
- tr->langopts.dotless_i = 1;
- tr->langopts.param[LOPT_SUFFIX] = 1;
- if (name2 == L('a', 'z'))
- tr->langopts.numbers = NUM_SINGLE_STRESS | NUM_DECIMAL_COMMA | NUM_ALLOW_SPACE | NUM_OMIT_1_HUNDRED | NUM_OMIT_1_THOUSAND | NUM_DFRACTION_2;
- else
- tr->langopts.numbers = NUM_SINGLE_STRESS | NUM_DECIMAL_COMMA | NUM_OMIT_1_HUNDRED | NUM_OMIT_1_THOUSAND | NUM_DFRACTION_2;
- tr->langopts.max_initial_consonants = 2;
- }
- break;
- case L('t', 't'): // Tatar
- {
- SetCyrillicLetters(tr);
- SetupTranslator(tr, stress_lengths_fr, stress_amps_fr);
- tr->langopts.stress_rule = STRESSPOSN_1R; // stress on final syllable
- tr->langopts.stress_flags = S_NO_AUTO_2; // no automatic secondary stress
- tr->langopts.numbers = NUM_SINGLE_STRESS | NUM_DECIMAL_COMMA | NUM_OMIT_1_HUNDRED | NUM_OMIT_1_THOUSAND | NUM_DFRACTION_4;
- }
- break;
- case L('u', 'k'): // Ukrainian
- {
- SetCyrillicLetters(tr);
- tr->langopts.param[LOPT_UNPRONOUNCABLE] = 0x432; // [v] don't count this character at start of word
- }
- break;
- case L('u', 'r'): // Urdu
- case L('s', 'd'): // Sindhi
- {
- tr->letter_bits_offset = OFFSET_ARABIC;
- tr->langopts.param[LOPT_UNPRONOUNCABLE] = 1; // disable check for unpronouncable words
- tr->langopts.numbers = NUM_SWAP_TENS;
- tr->langopts.break_numbers = 0x52a8; // for languages which have numbers for 100,000 and 100,00,000, eg Hindi
- }
- break;
- case L('v', 'i'): // Vietnamese
- {
- static const short stress_lengths_vi[8] = { 150, 150, 180, 180, 210, 230, 230, 240 };
- static const unsigned char stress_amps_vi[] = { 16, 16, 16, 16, 22, 22, 22, 22 };
- static wchar_t vowels_vi[] = {
- 0x61, 0xe0, 0xe1, 0x1ea3, 0xe3, 0x1ea1, // a
- 0x103, 0x1eb1, 0x1eaf, 0x1eb3, 0x1eb5, 0x1eb7, // ă
- 0xe2, 0x1ea7, 0x1ea5, 0x1ea9, 0x1eab, 0x1ead, // â
- 0x65, 0xe8, 0xe9, 0x1ebb, 0x1ebd, 0x1eb9, // e
- 0xea, 0x1ec1, 0x1ebf, 0x1ec3, 0x1ec5, 0x1ec7, // i
- 0x69, 0xec, 0xed, 0x1ec9, 0x129, 0x1ecb, // i
- 0x6f, 0xf2, 0xf3, 0x1ecf, 0xf5, 0x1ecd, // o
- 0xf4, 0x1ed3, 0x1ed1, 0x1ed5, 0x1ed7, 0x1ed9, // ô
- 0x1a1, 0x1edd, 0x1edb, 0x1edf, 0x1ee1, 0x1ee3, // ơ
- 0x75, 0xf9, 0xfa, 0x1ee7, 0x169, 0x1ee5, // u
- 0x1b0, 0x1eeb, 0x1ee9, 0x1eed, 0x1eef, 0x1ef1, // ư
- 0x79, 0x1ef3, 0xfd, 0x1ef7, 0x1ef9, 0x1ef5, // y
- 0
- };
- SetupTranslator(tr, stress_lengths_vi, stress_amps_vi);
- tr->langopts.length_mods0 = tr->langopts.length_mods; // don't lengthen vowels in the last syllable
- tr->langopts.stress_rule = STRESSPOSN_1L;
- tr->langopts.word_gap = 0x21; // length of a final vowel is less dependent on the next consonant, don't merge consonant with next word
- tr->letter_groups[0] = tr->letter_groups[7] = vowels_vi;
- tr->langopts.tone_language = 1; // Tone language, use CalcPitches_Tone() rather than CalcPitches()
- tr->langopts.unstressed_wd1 = 2;
- tr->langopts.numbers = NUM_DECIMAL_COMMA | NUM_HUNDRED_AND_DIGIT | NUM_DFRACTION_4 | NUM_ZERO_HUNDRED;
- }
- break;
- case L('w', 'o'):
- tr->langopts.stress_rule = STRESSPOSN_1L;
- tr->langopts.numbers = NUM_AND_UNITS | NUM_HUNDRED_AND | NUM_OMIT_1_HUNDRED | NUM_OMIT_1_THOUSAND | NUM_SINGLE_STRESS;
- break;
- case L3('c', 'm', 'n'): // no break, just go to 'zh' case
- case L3('z', 'h', 'y'): // just go to 'zh' case
- case L('z','h'):
- {
- static const short stress_lengths_zh[8] = { 230, 150, 230, 230, 230, 0, 240, 250 }; // 1=tone5. end-of-sentence, 6=tone 1&4, 7=tone 2&3
- static const unsigned char stress_amps_zh[] = { 22, 16, 22, 22, 22, 22, 22, 22 };
- SetupTranslator(tr, stress_lengths_zh, stress_amps_zh);
- tr->langopts.stress_rule = STRESSPOSN_1R; // stress on final syllable of a "word"
- tr->langopts.stress_flags = S_NO_DIM; // don't automatically set diminished stress (may be set in the intonation module)
- tr->langopts.vowel_pause = 0;
- tr->langopts.tone_language = 1; // Tone language, use CalcPitches_Tone() rather than CalcPitches()
- tr->langopts.length_mods0 = tr->langopts.length_mods; // don't lengthen vowels in the last syllable
- tr->langopts.tone_numbers = 1; // a number after letters indicates a tone number (eg. pinyin or jyutping)
- tr->langopts.ideographs = 1;
- tr->langopts.our_alphabet = 0x3100;
- tr->langopts.word_gap = 0x21; // length of a final vowel is less dependent on the next consonant, don't merge consonant with next word
- if (name2 == L3('z', 'h', 'y')) {
- tr->langopts.textmode = 1;
- tr->langopts.listx = 1; // compile zh_listx after zh_list
- tr->langopts.numbers = 1;
- tr->langopts.numbers2 = NUM2_ZERO_TENS;
- tr->langopts.break_numbers = 0x00018;
- }
- break;
- }
- default:
- tr->langopts.param[LOPT_UNPRONOUNCABLE] = 1; // disable check for unpronouncable words
- break;
- }
- tr->translator_name = name2;
- ProcessLanguageOptions(&tr->langopts);
- return tr;
- }
- void ProcessLanguageOptions(LANGUAGE_OPTIONS *langopts)
- {
- if (langopts->numbers & NUM_DECIMAL_COMMA) {
- // use . and ; for thousands and decimal separators
- langopts->thousands_sep = '.';
- langopts->decimal_sep = ',';
- }
- if (langopts->numbers & NUM_THOUS_SPACE)
- langopts->thousands_sep = 0; // don't allow thousands separator, except space
- }
- static void Translator_Russian(Translator *tr)
- {
- static const unsigned char stress_amps_ru[] = { 16, 16, 18, 18, 20, 24, 24, 22 };
- static const short stress_lengths_ru[8] = { 150, 140, 220, 220, 0, 0, 260, 280 };
- static const char ru_ivowels2[] = { 0x2c, 0x15, 0x18, 0x2e, 0x2f, 0 }; // add more vowels to letter group Y (iotated vowels & soft-sign)
- SetupTranslator(tr, stress_lengths_ru, stress_amps_ru);
- SetCyrillicLetters(tr);
- SetLetterBits(tr, 6, ru_ivowels2);
- tr->langopts.param[LOPT_UNPRONOUNCABLE] = 0x432; // [v] don't count this character at start of word
- tr->langopts.param[LOPT_REGRESSIVE_VOICING] = 1;
- tr->langopts.param[LOPT_REDUCE] = 2;
- tr->langopts.stress_rule = 5;
- tr->langopts.stress_flags = S_NO_AUTO_2;
- tr->langopts.numbers = NUM_DECIMAL_COMMA | NUM_OMIT_1_HUNDRED;
- tr->langopts.numbers2 = 0x2 + NUM2_THOUSANDS_VAR1; // variant numbers before thousands
- tr->langopts.phoneme_change = 1;
- tr->langopts.testing = 2;
- }
|