eSpeak NG is an open source speech synthesizer that supports more than hundred languages and accents.
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

phonemelist.cpp 17KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633
  1. /***************************************************************************
  2. * Copyright (C) 2005 to 2013 by Jonathan Duddington *
  3. * email: [email protected] *
  4. * *
  5. * This program is free software; you can redistribute it and/or modify *
  6. * it under the terms of the GNU General Public License as published by *
  7. * the Free Software Foundation; either version 3 of the License, or *
  8. * (at your option) any later version. *
  9. * *
  10. * This program is distributed in the hope that it will be useful, *
  11. * but WITHOUT ANY WARRANTY; without even the implied warranty of *
  12. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the *
  13. * GNU General Public License for more details. *
  14. * *
  15. * You should have received a copy of the GNU General Public License *
  16. * along with this program; if not, see: *
  17. * <http://www.gnu.org/licenses/>. *
  18. ***************************************************************************/
  19. #include "StdAfx.h"
  20. #include <stdio.h>
  21. #include <stdlib.h>
  22. #include <string.h>
  23. #include "speak_lib.h"
  24. #include "speech.h"
  25. #include "phoneme.h"
  26. #include "synthesize.h"
  27. #include "translate.h"
  28. const unsigned char pause_phonemes[8] = {0, phonPAUSE_VSHORT, phonPAUSE_SHORT, phonPAUSE, phonPAUSE_LONG, phonGLOTTALSTOP, phonPAUSE_LONG, phonPAUSE_LONG};
  29. extern int n_ph_list2;
  30. extern PHONEME_LIST2 ph_list2[N_PHONEME_LIST]; // first stage of text->phonemes
  31. static int SubstitutePhonemes(Translator *tr, PHONEME_LIST *plist_out)
  32. {//===================================================================
  33. // Copy the phonemes list and perform any substitutions that are required for the
  34. // current voice
  35. int ix;
  36. int k;
  37. int replace_flags;
  38. int n_plist_out = 0;
  39. int word_end;
  40. PHONEME_LIST2 *plist2;
  41. PHONEME_TAB *next=NULL;
  42. for(ix=0; (ix < n_ph_list2) && (n_plist_out < N_PHONEME_LIST); ix++)
  43. {
  44. plist2 = &ph_list2[ix];
  45. // don't do any substitution if the language has been temporarily changed
  46. if(!(plist2->synthflags & SFLAG_SWITCHED_LANG))
  47. {
  48. if(ix < (n_ph_list2 -1))
  49. next = phoneme_tab[ph_list2[ix+1].phcode];
  50. word_end = 0;
  51. if((plist2+1)->sourceix || ((next != 0) && (next->type == phPAUSE)))
  52. word_end = 1; // this phoneme is the end of a word
  53. // check whether a Voice has specified that we should replace this phoneme
  54. for(k=0; k<n_replace_phonemes; k++)
  55. {
  56. if(plist2->phcode == replace_phonemes[k].old_ph)
  57. {
  58. replace_flags = replace_phonemes[k].type;
  59. if((replace_flags & 1) && (word_end == 0))
  60. continue; // this replacement only occurs at the end of a word
  61. if((replace_flags & 2) && ((plist2->stresslevel & 0x7) > 3))
  62. continue; // this replacement doesn't occur in stressed syllables
  63. if((replace_flags & 4) && (plist2->sourceix == 0))
  64. continue; // this replacement only occurs at the start of a word
  65. // substitute the replacement phoneme
  66. plist2->phcode = replace_phonemes[k].new_ph;
  67. if((plist2->stresslevel > 1) && (phoneme_tab[plist2->phcode]->phflags & phUNSTRESSED))
  68. plist2->stresslevel = 0; // the replacement must be unstressed
  69. break;
  70. }
  71. }
  72. if(plist2->phcode == 0)
  73. {
  74. continue; // phoneme has been replaced by NULL, so don't copy it
  75. }
  76. }
  77. // copy phoneme into the output list
  78. memcpy(&plist_out[n_plist_out],plist2,sizeof(PHONEME_LIST2));
  79. plist_out[n_plist_out].ph = phoneme_tab[plist2->phcode];
  80. plist_out[n_plist_out].type = plist_out[n_plist_out].ph->type;
  81. n_plist_out++;
  82. }
  83. return(n_plist_out);
  84. } // end of SubstitutePhonemes
  85. void MakePhonemeList(Translator *tr, int post_pause, int start_sentence)
  86. {//=====================================================================
  87. int ix=0;
  88. int j;
  89. int insert_ph = 0;
  90. PHONEME_LIST *phlist;
  91. PHONEME_TAB *ph;
  92. PHONEME_TAB *next, *next2;
  93. int unstress_count = 0;
  94. int word_stress = 0;
  95. int current_phoneme_tab;
  96. int max_stress;
  97. int voicing;
  98. int regression;
  99. int end_sourceix;
  100. int alternative;
  101. int delete_count;
  102. PHONEME_DATA phdata;
  103. int n_ph_list3;
  104. PHONEME_LIST *plist3;
  105. PHONEME_LIST *plist3_inserted = NULL;
  106. PHONEME_LIST ph_list3[N_PHONEME_LIST];
  107. PHONEME_LIST2 *plist2;
  108. WORD_PH_DATA worddata;
  109. memset(&worddata, 0, sizeof(worddata));
  110. plist2 = ph_list2;
  111. phlist = phoneme_list;
  112. end_sourceix = plist2[n_ph_list2-1].sourceix;
  113. // is the last word of the clause unstressed ?
  114. max_stress = 0;
  115. for(j = n_ph_list2-3; j>=0; j--)
  116. {
  117. // start with the last phoneme (before the terminating pauses) and move backwards
  118. if((plist2[j].stresslevel & 0x7f) > max_stress)
  119. max_stress = plist2[j].stresslevel & 0x7f;
  120. if(plist2[j].sourceix != 0)
  121. break;
  122. }
  123. if(max_stress < 4)
  124. {
  125. // the last word is unstressed, look for a previous word that can be stressed
  126. while(--j >= 0)
  127. {
  128. if(plist2[j].synthflags & SFLAG_PROMOTE_STRESS) // dictionary flags indicated that this stress can be promoted
  129. {
  130. plist2[j].stresslevel = 4; // promote to stressed
  131. break;
  132. }
  133. if(plist2[j].stresslevel >= 4)
  134. {
  135. // found a stressed syllable, so stop looking
  136. break;
  137. }
  138. }
  139. }
  140. // look for switch of phoneme tables
  141. delete_count = 0;
  142. current_phoneme_tab = tr->phoneme_tab_ix;
  143. for(j = 0; j < n_ph_list2; j++)
  144. {
  145. if(current_phoneme_tab != tr->phoneme_tab_ix)
  146. {
  147. plist2[j].synthflags |= SFLAG_SWITCHED_LANG;
  148. }
  149. if(delete_count > 0)
  150. {
  151. memcpy(&plist2[j-delete_count], &plist2[j], sizeof(plist2[0]));
  152. }
  153. if(plist2[j].phcode == phonSWITCH)
  154. {
  155. if((!(plist2[j].synthflags & SFLAG_EMBEDDED)) && (
  156. (plist2[j].tone_ph == current_phoneme_tab) ||
  157. (plist2[j+1].phcode == phonSWITCH) ||
  158. ((plist2[j+1].phcode == phonPAUSE) && (plist2[j+2].phcode == phonSWITCH))
  159. ))
  160. {
  161. // delete this phonSWITCH if it's switching to the current phoneme table, or
  162. // delete this phonSWITCH if its followed by another phonSWITCH
  163. delete_count++;
  164. }
  165. else
  166. {
  167. current_phoneme_tab = plist2[j].tone_ph;
  168. }
  169. }
  170. }
  171. n_ph_list2 -= delete_count;
  172. if((regression = tr->langopts.param[LOPT_REGRESSIVE_VOICING]) != 0)
  173. {
  174. // set consonant clusters to all voiced or all unvoiced
  175. // Regressive
  176. int type;
  177. int stop_propagation = 0;
  178. voicing = 0;
  179. for(j=n_ph_list2-1; j>=0; j--)
  180. {
  181. ph = phoneme_tab[plist2[j].phcode];
  182. if(ph == NULL)
  183. continue;
  184. if(plist2[j].synthflags & SFLAG_SWITCHED_LANG)
  185. {
  186. stop_propagation = 0;
  187. voicing = 0;
  188. if(regression & 0x100)
  189. voicing = 1; // word-end devoicing
  190. continue;
  191. }
  192. type = ph->type;
  193. if(regression & 0x2)
  194. {
  195. // [v] amd [v;] don't cause regression, or [R^]
  196. if(((ph->mnemonic & 0xff) == 'v') || ((ph->mnemonic & 0xff)== 'R'))
  197. {
  198. stop_propagation = 1;
  199. if(regression & 0x10)
  200. voicing = 0;
  201. }
  202. }
  203. if((type==phSTOP) || type==(phFRICATIVE))
  204. {
  205. if((voicing==0) && (regression & 0xf))
  206. {
  207. voicing = 1;
  208. }
  209. else if((voicing==2) && (ph->end_type != 0)) // use end_type field for voicing_switch for consonants
  210. {
  211. plist2[j].phcode = ph->end_type; // change to voiced equivalent
  212. }
  213. }
  214. else if((type==phVSTOP) || type==(phVFRICATIVE))
  215. {
  216. if((voicing==0) && (regression & 0xf))
  217. {
  218. voicing = 2;
  219. }
  220. else if((voicing==1) && (ph->end_type != 0))
  221. {
  222. plist2[j].phcode = ph->end_type; // change to unvoiced equivalent
  223. }
  224. }
  225. else
  226. {
  227. if(regression & 0x8)
  228. {
  229. // LANG=Polish, propagate through liquids and nasals
  230. if((type == phPAUSE) || (type == phVOWEL))
  231. voicing = 0;
  232. }
  233. else
  234. {
  235. voicing = 0;
  236. }
  237. }
  238. if(stop_propagation)
  239. {
  240. voicing = 0;
  241. stop_propagation = 0;
  242. }
  243. if(plist2[j].sourceix)
  244. {
  245. if(regression & 0x04)
  246. {
  247. // stop propagation at a word boundary
  248. voicing = 0;
  249. }
  250. if(regression & 0x100)
  251. {
  252. // devoice word-final consonants, unless propagating voiced
  253. if(voicing == 0)
  254. {
  255. voicing = 1;
  256. }
  257. }
  258. }
  259. }
  260. }
  261. n_ph_list3 = SubstitutePhonemes(tr,ph_list3) - 2;
  262. for(j=0; (j < n_ph_list3) && (ix < N_PHONEME_LIST-3);)
  263. {
  264. if(ph_list3[j].sourceix)
  265. {
  266. // start of a word
  267. int k;
  268. int nextw;
  269. word_stress = 0;
  270. // find the highest stress level in this word
  271. for(nextw=j; nextw < n_ph_list3;)
  272. {
  273. if(ph_list3[nextw].stresslevel > word_stress)
  274. word_stress = ph_list3[nextw].stresslevel;
  275. nextw++;
  276. if(ph_list3[nextw].sourceix)
  277. break; // start of the next word
  278. }
  279. for(k=j; k<nextw; k++)
  280. {
  281. ph_list3[k].wordstress = word_stress;
  282. }
  283. j = nextw;
  284. }
  285. else
  286. {
  287. j++;
  288. }
  289. }
  290. // transfer all the phonemes of the clause into phoneme_list
  291. ph = phoneme_tab[phonPAUSE];
  292. ph_list3[0].ph = ph;
  293. for(j=0; insert_ph || ((j < n_ph_list3) && (ix < N_PHONEME_LIST-3)); j++)
  294. {
  295. plist3 = &ph_list3[j];
  296. if(insert_ph != 0)
  297. {
  298. // we have a (linking) phoneme which we need to insert here
  299. next = phoneme_tab[plist3->phcode]; // this phoneme, i.e. after the insert
  300. // re-use the previous entry for the inserted phoneme.
  301. // That's OK because we don't look backwards from plist3 *** but CountVowelPosiion() and isAfterStress does !!!
  302. j--;
  303. plist3 = plist3_inserted = &ph_list3[j];
  304. if(j > 0)
  305. {
  306. memcpy(&plist3[-1], &plist3[0], sizeof(*plist3));
  307. }
  308. memset(&plist3[0], 0, sizeof(*plist3));
  309. plist3->phcode = insert_ph;
  310. ph = phoneme_tab[insert_ph];
  311. plist3->ph = ph;
  312. insert_ph = 0;
  313. }
  314. else
  315. {
  316. // otherwise get the next phoneme from the list
  317. ph = phoneme_tab[plist3->phcode];
  318. plist3[0].ph = ph;
  319. if(plist3->phcode == phonSWITCH)
  320. {
  321. // change phoneme table
  322. SelectPhonemeTable(plist3->tone_ph);
  323. }
  324. next = phoneme_tab[plist3[1].phcode]; // the phoneme after this one
  325. plist3[1].ph = next;
  326. }
  327. if(ph == NULL) continue;
  328. InterpretPhoneme(tr, 0x100, plist3, &phdata, &worddata);
  329. if((alternative = phdata.pd_param[pd_INSERTPHONEME]) > 0)
  330. {
  331. // PROBLEM: if we insert a phoneme before a vowel then we loose the stress.
  332. PHONEME_TAB *ph2;
  333. ph2 = ph;
  334. insert_ph = plist3->phcode;
  335. ph = phoneme_tab[alternative];
  336. plist3->ph = ph;
  337. plist3->phcode = alternative;
  338. if(ph->type == phVOWEL)
  339. {
  340. plist3->synthflags |= SFLAG_SYLLABLE;
  341. if(ph2->type != phVOWEL)
  342. plist3->stresslevel = 0; // change from non-vowel to vowel, make sure it's unstressed
  343. }
  344. else
  345. plist3->synthflags &= ~SFLAG_SYLLABLE;
  346. // re-interpret the changed phoneme
  347. // But it doesn't obey a second ChangePhoneme()
  348. InterpretPhoneme(tr, 0x100, plist3, &phdata, &worddata);
  349. }
  350. if((alternative = phdata.pd_param[pd_CHANGEPHONEME]) > 0)
  351. {
  352. PHONEME_TAB *ph2;
  353. ph2 = ph;
  354. ph = phoneme_tab[alternative];
  355. plist3->ph = ph;
  356. plist3->phcode = alternative;
  357. if(alternative == 1)
  358. continue; // NULL phoneme, discard
  359. if(ph->type == phVOWEL)
  360. {
  361. plist3->synthflags |= SFLAG_SYLLABLE;
  362. if(ph2->type != phVOWEL)
  363. plist3->stresslevel = 0; // change from non-vowel to vowel, make sure it's unstressed
  364. }
  365. else
  366. plist3->synthflags &= ~SFLAG_SYLLABLE;
  367. // re-interpret the changed phoneme
  368. // But it doesn't obey a second ChangePhoneme()
  369. InterpretPhoneme(tr, 0x100, plist3, &phdata, &worddata);
  370. }
  371. if(ph->type == phVOWEL)
  372. {
  373. PHONEME_LIST *p;
  374. // Check for consecutive unstressed syllables, even across word boundaries.
  375. // Do this after changing phonemes according to stress level.
  376. if(plist3->stresslevel <= 1)
  377. {
  378. // an unstressed vowel
  379. unstress_count++;
  380. if(tr->langopts.stress_flags & 0x08)
  381. {
  382. // change sequences of consecutive unstressed vowels in unstressed words to diminished stress (TEST)
  383. for(p=plist3+1; p->type != phPAUSE; p++)
  384. {
  385. if(p->type == phVOWEL)
  386. {
  387. if(p->stresslevel <= 1)
  388. {
  389. if(plist3->wordstress < 4)
  390. plist3->stresslevel = 0;
  391. if(p->wordstress < 4)
  392. p->stresslevel = 0;
  393. }
  394. break;
  395. }
  396. }
  397. }
  398. else
  399. {
  400. if((unstress_count > 1) && ((unstress_count & 1)==0))
  401. {
  402. // in a sequence of unstressed syllables, reduce alternate syllables to 'diminished'
  403. // stress. But not for the last phoneme of a stressed word
  404. if((tr->langopts.stress_flags & S_NO_DIM) || ((word_stress > 3) && ((plist3+1)->sourceix!=0)))
  405. {
  406. // An unstressed final vowel of a stressed word
  407. unstress_count=1; // try again for next syllable
  408. }
  409. else
  410. {
  411. plist3->stresslevel = 0; // change stress to 'diminished'
  412. }
  413. }
  414. }
  415. }
  416. else
  417. {
  418. unstress_count = 0;
  419. }
  420. }
  421. if((plist3+1)->synthflags & SFLAG_LENGTHEN)
  422. {
  423. static char types_double[] = {phFRICATIVE,phVFRICATIVE,phNASAL,phLIQUID,0};
  424. if(strchr(types_double,next->type))
  425. {
  426. // lengthen this consonant by doubling it
  427. insert_ph = next->code;
  428. (plist3+1)->synthflags ^= SFLAG_LENGTHEN;
  429. }
  430. }
  431. if((plist3+1)->sourceix != 0)
  432. {
  433. int x;
  434. if(tr->langopts.vowel_pause && (ph->type != phPAUSE))
  435. {
  436. if((ph->type != phVOWEL) && (tr->langopts.vowel_pause & 0x200))
  437. {
  438. // add a pause after a word which ends in a consonant
  439. insert_ph = phonPAUSE_NOLINK;
  440. }
  441. if(next->type == phVOWEL)
  442. {
  443. if((x = tr->langopts.vowel_pause & 0x0c) != 0)
  444. {
  445. // break before a word which starts with a vowel
  446. if(x == 0xc)
  447. insert_ph = phonPAUSE_NOLINK;
  448. else
  449. insert_ph = phonPAUSE_VSHORT;
  450. }
  451. if((ph->type == phVOWEL) && ((x = tr->langopts.vowel_pause & 0x03) != 0))
  452. {
  453. // adjacent vowels over a word boundary
  454. if(x == 2)
  455. insert_ph = phonPAUSE_SHORT;
  456. else
  457. insert_ph = phonPAUSE_VSHORT;
  458. }
  459. if(((plist3+1)->stresslevel >= 4) && (tr->langopts.vowel_pause & 0x100))
  460. {
  461. // pause before a words which starts with a stressed vowel
  462. insert_ph = phonPAUSE_SHORT;
  463. }
  464. }
  465. }
  466. if(plist3 != plist3_inserted)
  467. {
  468. if((x = (tr->langopts.word_gap & 0x7)) != 0)
  469. {
  470. if((x > 1) || ((insert_ph != phonPAUSE_SHORT) && (insert_ph != phonPAUSE_NOLINK)))
  471. {
  472. // don't reduce the pause
  473. insert_ph = pause_phonemes[x];
  474. }
  475. }
  476. if(option_wordgap > 0)
  477. {
  478. insert_ph = phonPAUSE_LONG;
  479. }
  480. }
  481. }
  482. next2 = phoneme_tab[plist3[2].phcode];
  483. plist3[2].ph = next2;
  484. if((insert_ph == 0) && (phdata.pd_param[pd_APPENDPHONEME] != 0))
  485. {
  486. insert_ph = phdata.pd_param[pd_APPENDPHONEME];
  487. }
  488. if(ph->phflags & phVOICED)
  489. {
  490. // check that a voiced consonant is preceded or followed by a vowel or liquid
  491. // and if not, add a short schwa
  492. // not yet implemented
  493. }
  494. phlist[ix].ph = ph;
  495. phlist[ix].type = ph->type;
  496. phlist[ix].env = PITCHfall; // default, can be changed in the "intonation" module
  497. phlist[ix].synthflags = plist3->synthflags;
  498. phlist[ix].stresslevel = plist3->stresslevel & 0xf;
  499. phlist[ix].wordstress = plist3->wordstress;
  500. phlist[ix].tone_ph = plist3->tone_ph;
  501. phlist[ix].sourceix = 0;
  502. phlist[ix].phcode = ph->code;
  503. if(plist3->sourceix != 0)
  504. {
  505. phlist[ix].sourceix = plist3->sourceix;
  506. phlist[ix].newword = 1; // this phoneme is the start of a word
  507. if(start_sentence)
  508. {
  509. phlist[ix].newword = 5; // start of sentence + start of word
  510. start_sentence = 0;
  511. }
  512. }
  513. else
  514. {
  515. phlist[ix].newword = 0;
  516. }
  517. // phlist[ix].length = ph->std_length;
  518. phlist[ix].length = phdata.pd_param[i_SET_LENGTH]*2;
  519. if((ph->code == phonPAUSE_LONG) && (option_wordgap > 0))
  520. {
  521. phlist[ix].ph = phoneme_tab[phonPAUSE_SHORT];
  522. phlist[ix].length = option_wordgap*14; // 10mS per unit at the default speed
  523. }
  524. if(ph->type==phVOWEL || ph->type==phLIQUID || ph->type==phNASAL || ph->type==phVSTOP || ph->type==phVFRICATIVE)
  525. {
  526. phlist[ix].length = 128; // length_mod
  527. phlist[ix].env = PITCHfall;
  528. }
  529. phlist[ix].prepause = 0;
  530. phlist[ix].amp = 20; // default, will be changed later
  531. phlist[ix].pitch1 = 255;
  532. phlist[ix].pitch2 = 255;
  533. ix++;
  534. }
  535. phlist[ix].newword = 2; // end of clause
  536. phlist[ix].phcode = phonPAUSE;
  537. phlist[ix].type = phPAUSE; // terminate with 2 Pause phonemes
  538. phlist[ix].length = post_pause; // length of the pause, depends on the punctuation
  539. phlist[ix].sourceix = end_sourceix;
  540. phlist[ix].synthflags = 0;
  541. phlist[ix++].ph = phoneme_tab[phonPAUSE];
  542. phlist[ix].phcode = phonPAUSE;
  543. phlist[ix].type = phPAUSE;
  544. phlist[ix].length = 0;
  545. phlist[ix].sourceix=0;
  546. phlist[ix].synthflags = 0;
  547. phlist[ix++].ph = phoneme_tab[phonPAUSE_SHORT];
  548. n_phoneme_list = ix;
  549. } // end of MakePhonemeList