/*************************************************************************** * Copyright (C) 2006 to 2011 by Jonathan Duddington * * email: jonsd@users.sourceforge.net * * * * 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, write see: * * . * ***************************************************************************/ #include "wx/wx.h" #include #include "wx/filename.h" #include "wx/sound.h" #include "wx/dir.h" #include #include "speak_lib.h" #include "main.h" #include "speech.h" #include "phoneme.h" #include "synthesize.h" #include "voice.h" #include "spect.h" #include "translate.h" #include "options.h" extern char word_phonemes[N_WORD_PHONEMES]; // a word translated into phoneme codes extern int __cdecl string_sorter(char **a, char **b); //****************************************************************************************************** FILE *f_wavtest = NULL; FILE *f_events = NULL; FILE *OpenWaveFile3(const char *path) /***********************************/ { int *p; FILE *f; static unsigned char wave_hdr[44] = { 'R','I','F','F',0,0,0,0,'W','A','V','E','f','m','t',' ', 0x10,0,0,0,1,0,1,0, 9,0x3d,0,0,0x12,0x7a,0,0, 2,0,0x10,0,'d','a','t','a', 0,0,0,0 }; if(path == NULL) return(NULL); // set the sample rate in the header p = (int *)(&wave_hdr[24]); p[0] = samplerate; p[1] = samplerate * 2; f = fopen(path,"wb"); if(f != NULL) { fwrite(wave_hdr,1,sizeof(wave_hdr),f); } return(f); } // end of OpenWaveFile void CloseWaveFile3(FILE *f) /*************************/ { unsigned int pos; static int value; if(f == NULL) return; fflush(f); pos = ftell(f); value = pos - 8; fseek(f,4,SEEK_SET); fwrite(&value,4,1,f); value = samplerate; fseek(f,24,SEEK_SET); fwrite(&value,4,1,f); value = samplerate*2; fseek(f,28,SEEK_SET); fwrite(&value,4,1,f); value = pos - 44; fseek(f,40,SEEK_SET); fwrite(&value,4,1,f); fclose(f); } // end of CloseWaveFile3 int TestUriCallback(int type, const char *uri, const char *base) {//============================================================= if(strcmp(uri,"hello")==0) return(1); return(0); } int TestSynthCallback(short *wav, int numsamples, espeak_EVENT *events) {//==================================================================== int type; f_events = fopen("/home/jsd1/speechdata/text/events","a"); fprintf(f_events,"--\n"); while((type = events->type) != 0) { fprintf(f_events,"%2d (%4d %4ld) %5d %5d (%3d) ",type,events->unique_identifier,(long)events->user_data,events->audio_position,events->text_position,events->length); if((type==3) || (type==4)) fprintf(f_events,"'%s'\n",events->id.name); else if(type==espeakEVENT_PHONEME) { char buf[10]; // fprintf(f_events,"[%s]\n",WordToString(events->id.number)); //old version, only 4 characters bytes memcpy(buf, events->id.string, 8); buf[8] = 0; fprintf(f_events,"[%s]\n", buf); } else fprintf(f_events,"%d\n",events->id.number); events++; } if((wav == NULL) && (f_wavtest != NULL)) { fprintf(f_events,"Finished\n"); CloseWaveFile3(f_wavtest); f_wavtest = NULL; } fclose(f_events); if(f_wavtest == NULL) return(0); fwrite(wav,numsamples*2,1,f_wavtest); return(0); } //****************************************************************************************************** #ifdef deleted static int RuLex_sorter(char **a, char **b) {//======================================= char *pa, *pb; int xa, xb; int ix; pa = *a; pb = *b; xa = strlen(pa)-1; xb = strlen(pb)-1; while((xa >= 0) && (xb >= 0)) { if((ix = (pa[xa] - pb[xb])) != 0) return(ix); xa--; xb--; } return(pa - pb); } /* end of strcmp2 */ #endif static const unsigned short KOI8_R[0x60] = { 0x2550, 0x2551, 0x2552, 0x0451, 0x2553, 0x2554, 0x2555, 0x2556, // a0 0x2557, 0x2558, 0x2559, 0x255a, 0x255b, 0x255c, 0x255d, 0x255e, // a8 0x255f, 0x2560, 0x2561, 0x0401, 0x2562, 0x2563, 0x2564, 0x2565, // b0 0x2566, 0x2567, 0x2568, 0x2569, 0x256a, 0x256b, 0x256c, 0x00a9, // b8 0x044e, 0x0430, 0x0431, 0x0446, 0x0434, 0x0435, 0x0444, 0x0433, // c0 0x0445, 0x0438, 0x0439, 0x043a, 0x043b, 0x043c, 0x043d, 0x043e, // c8 0x043f, 0x044f, 0x0440, 0x0441, 0x0442, 0x0443, 0x0436, 0x0432, // d0 0x044c, 0x044b, 0x0437, 0x0448, 0x044d, 0x0449, 0x0447, 0x044a, // d8 0x042e, 0x0410, 0x0411, 0x0426, 0x0414, 0x0415, 0x0424, 0x0413, // e0 0x0425, 0x0418, 0x0419, 0x041a, 0x041b, 0x041c, 0x041d, 0x041e, // e8 0x041f, 0x042f, 0x0420, 0x0421, 0x0422, 0x0423, 0x0416, 0x0412, // f0 0x042c, 0x042b, 0x0417, 0x0428, 0x042d, 0x0429, 0x0427, 0x042a, // f8 }; #define N_CHARS 34 #define PH(c1,c2) (c2<<8)+c1 // combine two characters into an integer for phoneme name static void DecodePhonemes2(const char *inptr, char *outptr) //=================================================== // Translate from internal phoneme codes into phoneme mnemonics // This version is for Lexicon_De() { unsigned char phcode; unsigned char c; unsigned int mnem; PHONEME_TAB *ph; const char *p; int ix; int j; int start; static const char *stress_chars = "==,,'* "; unsigned int replace_ph[] = {',',PH('@','-'),'W','3','y','A',PH('A',':'),'*',PH('_','!'),PH('_','|'),PH('O','I'),PH('Y',':'),PH('p','F'),PH('E','2'),0}; const char *replace_ph2[] = {NULL,NULL,"9","@r","Y","a:", "a:", "r", "?", "?", "OY", "2:", "pf" ,"E",NULL}; start = 1; for(ix=0; (phcode = inptr[ix]) != 0; ix++) { if(phcode == 255) continue; /* indicates unrecognised phoneme */ if((ph = phoneme_tab[phcode]) == NULL) continue; if((ph->type == phSTRESS) && (ph->std_length <= 4) && (ph->program == 0)) { if(ph->std_length > 2) *outptr++ = stress_chars[ph->std_length]; } else { mnem = ph->mnemonic; if(ph->type == phPAUSE) { if(start) continue; // omit initial [?] if(inptr[ix+1] == phonSCHWA_SHORT) continue; // omit [?] before [@-*] } start = 0; p = NULL; for(j=0;;j++) { if(replace_ph[j] == 0) break; if(mnem == replace_ph[j]) { p = replace_ph2[j]; if(p == NULL) mnem = 0; break; } } if(p != NULL) { while((c = *p++) != 0) { *outptr++ = c; } } else if(mnem != 0) { while((c = (mnem & 0xff)) != 0) { *outptr++ = c; mnem = mnem >> 8; } } } } *outptr = 0; /* string terminator */ } // end of DecodePhonemes2 void Lexicon_It(int pass) {//====================== // Reads a lexicon of pairs of words: normal spelling and spelling with accents // Creates file: dictsource/it_listx which includes corrections for stress position and [E]/[e] and [O]/[o] phonemes // Words which are still in error are listed in file: it_compare (in the directory of the lexicon file). int count=0; int matched=0; int ix; int c; char *p, *p2; int len; int vowel_ix; int stress_posn1; int stress_posn2; int stress_vowel1; int stress_vowel2; int use_phonemes; FILE *f_in; FILE *f_out; FILE *f_listx; FILE *f_list_in = NULL; int listx_count; long int displ; const char *alt_string; wxString str; static wxString fname_lex; char buf[200]; char word[80]; char word1[80]; char word2[80]; char word_stem[80]; char temp[80]; char phonemes[80]; char phonemes2[80]; char phonemes3[80]; char buf_out[200]; char buf_error[200]; char last_listx[200]; static const char *vowels1 = "aeiou"; static const char *vowels2 = "aeou"; static const char ex1[] = {'a',0xc3,0xac,0}; // aì static const char ex2[] = {'e',0xc3,0xac,0}; // eì static const char ex3[] = {0xc3,0xb9,'a',0}; // ùa static const char ex4[] = {0xc3,0xb9,'e',0}; // ùe static const char ex5[] = {0xc3,0xb9,'i',0}; // ùi static const char ex6[] = {0xc3,0xb9,'o',0}; // ùo static const char ex7[] = {'c',0xc3,0xac,'a',0}; // cìa static const char ex8[] = {'c',0xc3,0xac,'o',0}; // cìo static const char ex9[] = {'c',0xc3,0xac,'u',0}; // cìu static const char ex10[] = {'g','l',0xc3,0xac,0}; // glì static const char *exceptions[] = {ex1, ex2, ex3, ex4, ex5, ex6, ex7, ex8, ex9, ex10, NULL}; if(pass == 1) { fname_lex = wxFileSelector(_T("Italian Lexicon"),path_dir1,_T(""),_T(""),_T("*"),wxOPEN); } strcpy(buf,fname_lex.mb_str(wxConvLocal)); if((f_in = fopen(buf,"r")) == NULL) { wxLogError(_T("Can't read file ")+fname_lex); return; } path_dir1 = wxFileName(fname_lex).GetPath(); strcpy(buf_out, path_dir1.mb_str(wxConvLocal)); sprintf(buf, "%s/IT_errors", buf_out); if((f_out = fopen(buf,"w")) == NULL) { str = wxString(buf, wxConvLocal); wxLogError(_T("Can't write file: ") + str); return; } listx_count = 0; last_listx[0] = 0; if(pass == 1) { sprintf(buf,"%s/it_listx",path_dsource); remove(buf); CompileDictionary(path_dsource,"it",NULL,NULL,0); f_listx = fopen(buf,"w"); wxLogStatus(_T("Pass 1")); } else { CompileDictionary(path_dsource,"it",NULL,NULL,0); sprintf(buf,"%s/it_listx2",path_dsource); f_listx = fopen(buf,"w"); sprintf(buf,"%s/it_listx",path_dsource); if((f_list_in = fopen(buf,"r")) == NULL) { wxLogError(_T("Can't read file: it_listx")); return; } } if(f_listx == NULL) { wxLogError(_T("Can't write file: it_listx")); return; } LoadVoice("it",0); while(!feof(f_in)) { count++; if(fgets(buf,sizeof(buf),f_in) == NULL) break; if((p = strstr(buf,"//")) != NULL) *p = 0; if((sscanf(buf,"%s %s",word,temp)) < 2) continue; if(strlen(word) < 8) sprintf(buf_error,"%s\t\t%s\t",word,temp); else sprintf(buf_error,"%s\t%s",word,temp); sprintf(word1," %s ",word); // should we remove a vowel ending to produce a stem ? strcpy(word_stem, word); len = strlen(word) - 1; utf8_in(&c, temp); // if(iswlower(c)) { if((word[len] == 'a') && (strchr(vowels1, word[len-1]) == NULL)) word_stem[len] = 0; else if((word[len] == 'o') && (strchr(vowels2, word[len-1]) == NULL)) word_stem[len] = 0; } // convert word to lower-case word2[0] = ' '; for(ix=0, p=&word2[1];;) { ix += utf8_in(&c,&temp[ix]); c = towlower(c); p += utf8_out(c,p); if(c == 0) break; } strcat(word2," "); use_phonemes = 0; for(ix=0; ; ix++) { if(exceptions[ix] == NULL) break; if(strstr(word2, exceptions[ix]) != NULL) { // the word contains a string for which we must do a complete phoneme translation use_phonemes = 1; strcpy(word_stem, word); break; } } // translate TranslateWord(translator,&word1[1],0, NULL); DecodePhonemes(word_phonemes,phonemes); stress_posn1 = 0; stress_posn2 = 0; stress_vowel1 = 0; stress_vowel2 = 0; vowel_ix = 1; for(ix=0; ;ix++) { if((c = word_phonemes[ix]) == 0) break; if(c == phonSTRESS_P) { stress_posn1 = vowel_ix; stress_vowel1 = word_phonemes[ix+1]; } if((c != phonSCHWA_SHORT) && (phoneme_tab[c]->type == phVOWEL)) vowel_ix++; } TranslateWord(translator,&word2[1],0, NULL); DecodePhonemes(word_phonemes,phonemes2); vowel_ix = 1; for(ix=0; ;ix++) { if((c = word_phonemes[ix]) == 0) break; if(c == phonSTRESS_P) { stress_posn2 = vowel_ix; stress_vowel2 = word_phonemes[ix+1]; } if((c != phonSCHWA_SHORT) && (phoneme_tab[c]->type == phVOWEL)) vowel_ix++; } if(stress_posn2 == (vowel_ix-1)) { // stress is on the final vowel, don't renove it strcpy(word_stem, word); } if(pass == 1) { if(use_phonemes) { fprintf(f_listx,"%s ", word_stem); for(p = phonemes2; *p != 0; p++) { if(*p != ',') fputc(*p, f_listx); // omit secondary stress marks } fputc('\n',f_listx); } else if((stress_posn1 != stress_posn2) && (stress_posn1 > 0) && (stress_posn2 > 0)) { fprintf(f_listx,"%s $%d\n", word_stem, stress_posn2); } } // reduce [E] and [O] to [e] and [o] if not stressed for(ix=0; phonemes[ix] != 0; ix++) { if((phonemes[ix] == 'E') || (phonemes[ix] == 'O')) { if((pass == 2) || (ix==0) || (phonemes[ix-1] != '\'')) phonemes[ix] = tolower(phonemes[ix]); } } for(ix=0; phonemes2[ix] != 0; ix++) { if((phonemes2[ix] == 'E') || (phonemes2[ix] == 'O')) { if((pass == 2) || (ix==0) || (phonemes2[ix-1] != '\'')) phonemes2[ix] = tolower(phonemes2[ix]); } } if(strcmp(phonemes,phonemes2) == 0) { alt_string = NULL; if((pass == 2) && (stress_posn1 > 0) && (stress_posn2 > 0)) { if(((stress_vowel1 == PhonemeCode('E')) && (stress_vowel2 == PhonemeCode('e'))) || ((stress_vowel1 == PhonemeCode('O')) && (stress_vowel2 == PhonemeCode('o')))) { alt_string = " $alt2"; } else if(((stress_vowel1 == PhonemeCode('e')) && (stress_vowel2 == PhonemeCode('E'))) || ((stress_vowel1 == PhonemeCode('o')) && (stress_vowel2 == PhonemeCode('O')))) { alt_string = " $alt"; } if(alt_string != NULL) { while(!feof(f_list_in)) { displ = ftell(f_list_in); if(fgets(buf, sizeof(buf), f_list_in) == NULL) break; sscanf(buf, "%s", word1); if(strcmp(word1, word_stem) < 0) { sprintf(buf_out,"%s",buf); // copy it_listx from pass 1 until we reach the matching word } else { if(strcmp(word1, word_stem) == 0) { p = buf; while((*p != '\n') && (*p != 0)) *p++; *p = 0; sprintf(buf_out,"%s %s\n",buf,alt_string); // add $alt or $alt2 to the entry } else { sprintf(buf_out,"%s %s\n", word_stem, alt_string); // add a new word with $alt or $alt2 fseek(f_list_in, displ, SEEK_SET); } if(strcmp(buf_out, last_listx) != 0) { fprintf(f_listx, "%s", buf_out); listx_count++; strcpy(last_listx, buf_out); } break; } if(strcmp(buf_out, last_listx) != 0) { fprintf(f_listx, "%s", buf_out); listx_count++; strcpy(last_listx, buf_out); } } } } matched++; } else { // allow if the only difference is no primary stress p2 = phonemes2; p = phonemes3; while(*p2 != 0) { *p = *p2++; if((*p2 == ':') && (strchr("aeiouEO", *p) != NULL)) p2++; // change lone vowels to short by removing ':' if(*p == '\'') *p = ','; // change primary to secondary stress p++; } *p = 0; if(strcmp(phonemes, phonemes3) == 0) { matched++; } else { // still doesn't match, report this word fprintf(f_out,"%s\t%s\t%s\n",buf_error,phonemes,phonemes2); } } } if(pass == 2) { while(fgets(buf, sizeof(buf), f_list_in) != NULL) { if(strcmp(buf, last_listx) != 0) // check for duplicate entries { fprintf(f_listx, "%s", buf); // copy the remaining entries from pass 1 listx_count++; strcpy(last_listx, buf); } } fclose(f_list_in); } fclose(f_in); fclose(f_out); fclose(f_listx); if(pass == 2) { sprintf(buf,"%s/it_listx",path_dsource); remove(buf); sprintf(buf_out,"%s/it_listx2",path_dsource); rename(buf_out, buf); wxLogStatus(_T("Created file 'it_listx', entries=%d errors=%d total words=%d"),listx_count, count-matched, count); } else { wxLogStatus(_T("Pass 1, equal=%d different=%d"),matched,count-matched); } } // end of Lexicon_It void Lexicon_De() {//============== // Compare eSpeak's translation of German words with a pronunciation lexicon FILE *f_in; FILE *f_out; int ix; int c; int c2; char *p; int stress; int count=0; int start; int matched=0; int defer_stress = 0; char buf[200]; char word[80]; char word2[80]; char type[80]; char pronounce[80]; char pronounce2[80]; char phonemes[80]; char phonemes2[80]; static const char *vowels = "aeiouyAEIOUY29@"; wxString fname = wxFileSelector(_T("German Lexicon"),path_dir1,_T(""),_T(""),_T("*"),wxOPEN); strcpy(buf,fname.mb_str(wxConvLocal)); if((f_in = fopen(buf,"r")) == NULL) { wxLogError(_T("Can't read file ")+fname); return; } path_dir1 = wxFileName(fname).GetPath(); if((f_out = fopen("compare_de","w")) == NULL) { wxLogError(_T("Can't write file ")); return; } LoadVoice("de",0); word2[0] = ' '; while(!feof(f_in)) { count++; if(fgets(buf,sizeof(buf),f_in) == NULL) break; sscanf(buf,"%s %s %s",word,type,pronounce); // convert word to lower-case for(ix=0, p=&word2[1];;) { ix += utf8_in(&c,&word[ix]); c = towlower(c); p += utf8_out(c,p); if(c == 0) break; } strcpy(word,&word2[1]); strcat(&word2[1]," "); // remove | syllable boundaries stress=0; start=1; for(ix=0, p=pronounce2;;ix++) { c = pronounce[ix]; if(c == '\'') { stress=4; continue; } if(c == ',') { stress=3; continue; } if(c == '|') continue; if((c == '?') && start) continue; // omit initial [?] if(c == '<') { if((c2 = pronounce[ix+1]) == 'i') { defer_stress =1; #ifdef deleted if(stress == 4) { *p++ = 'i'; c =':'; } else #endif { c = 'I'; } ix++; } } start =0; if(defer_stress) { defer_stress = 0; } else if(stress && (strchr(vowels,c) != NULL)) { if(stress == 4) *p++ = '\''; if(stress == 3) *p++ = ','; stress = 0; } *p++ = c; if(c == 0) break; if(strchr("eiouy",c) && pronounce[ix+1] != ':') *p++ = ':'; // ensure [;] after these vowels } // translate TranslateWord(translator,&word2[1],0, NULL); DecodePhonemes2(word_phonemes,phonemes); // also need to change some phoneme names if(strcmp(phonemes,pronounce2) == 0) { matched++; } else { // remove secondary stress strcpy(phonemes2,phonemes); p = phonemes; for(ix=0; ;ix++) { if((c = phonemes2[ix]) != ',') *p++ = c; if(c == 0) break; } if(strcmp(phonemes,pronounce2) == 0) { matched++; } else { if(strlen(word) < 8) strcat(word,"\t"); fprintf(f_out,"%s\t%s\t%s\n",word,phonemes,pronounce2); } } } fclose(f_in); fclose(f_out); wxLogStatus(_T("Completed, equal=%d different=%d"),matched,count-matched); } // end of Lexicon_De extern int IsVowel(Translator *tr, int letter); void Lexicon_Test() {//================ int c1, c2, c3; char *p; int prev_c1=0; int prev_c2=0; int prev_c3 = 0; FILE *f_in; FILE *f_out; char buf[200]; wxString s_fname = wxFileSelector(_T("List of UTF-8 words with Combining Grave Accent U+300 to indicate stress"),path_dir1, _T(""),_T(""),_T("*"),wxOPEN); if(s_fname.IsEmpty()) return; strcpy(buf,s_fname.mb_str(wxConvLocal)); path_dir1 = wxFileName(s_fname).GetPath(); if((f_in = fopen(buf,"r")) == NULL) { wxLogError(_T("Can't read file: ") + wxString(buf,wxConvLocal)); return; } strcat(buf,"_1"); if((f_out = fopen(buf,"w")) == NULL) { wxLogError(_T("Can't write file: ") + wxString(buf,wxConvLocal)); fclose(f_in); return; } while(!feof(f_in)) { if((p = fgets(buf,sizeof(buf),f_in)) == NULL) break; if(buf[0] == 0) continue; p += utf8_in(&c1, p); p += utf8_in(&c2, p); p += utf8_in(&c3, p); c1 = towlower(c1); c2 = towlower(c2); c3 = towlower(c3); if(IsVowel(translator, c1)) continue; if(IsVowel(translator, c2)) continue; if((prev_c1 != c1) || (prev_c2 != c2) || ((prev_c3 != c3) && !IsVowel(translator,c3))) fputc('\n',f_out); prev_c1 = c1; prev_c2 = c2; prev_c3 = c3; fprintf(f_out,"%s",buf); } fclose(f_in); fclose(f_out); } // end of Lexicon_Test void Lexicon_Bg() {//============== // Bulgarian: compare stress markup in a list of words with lookup using bg_rules char *p; char *pw; char *pw1; int cc; int vcount; int lex_stress; int input_length; int n_words=0; int n_wrong=0; int n_out=0; int n_stress; int max_stress; int max_stress_posn; int stress_first; int done; PHONEME_TAB *ph; FILE *f_in; FILE *f_out; char word[80]; char word_in[80]; char phonemes[N_WORD_PHONEMES]; char buf[200]; char fname[sizeof(path_dsource)+20]; static unsigned short bg_vowels[] = {0x430, 0x435, 0x438, 0x43e, 0x443, 0x44a, 0x44d, 0x44e, 0x44f, 0x450, 0x451, 0x45d, 0}; if(gui_flag == 0) return; wxString s_fname = wxFileSelector(_T("List of UTF-8 words with Combining Grave Accent U+300 to indicate stress"),path_dir1, _T(""),_T(""),_T("*"),wxOPEN); if(s_fname.IsEmpty()) return; strcpy(buf,s_fname.mb_str(wxConvLocal)); path_dir1 = wxFileName(s_fname).GetPath(); if((f_in = fopen(buf,"r")) == NULL) { wxLogError(_T("Can't read file: ") + wxString(buf,wxConvLocal)); return; } input_length = GetFileLength(buf); sprintf(fname,"%s%c%s",path_dsource,PATHSEP,"bg_listx"); remove(fname); CompileDictionary(path_dsource,"bg",NULL,NULL,0); if((f_out = fopen(fname,"w")) == NULL) { wxLogError(_T("Can't write to: ")+wxString(fname,wxConvLocal)); fclose(f_in); return; } LoadVoice("bg",0); progress = new wxProgressDialog(_T("Lexicon"),_T(""),input_length); for(;;) { if((n_words & 0x3ff) == 0) { progress->Update(ftell(f_in)); } if(fgets(buf,sizeof(buf),f_in) == NULL) break; if(isspace2(buf[0])) continue; // convert from UTF-8 to Unicode word[0] = 0; word[1] = ' '; pw = &word[2]; pw1 = word_in; p = buf; while(*p == ' ') p++; vcount = 0; lex_stress = 0; n_stress = 0; stress_first = 0; // find the marked stress position for(;;) { p += utf8_in(&cc, p); if(iswspace(cc)) break; if(cc == 0xfeff) continue; // ignore UTF-8 indication if(cc == '`') cc = '\''; pw1 += utf8_out(towlower(cc), pw1); // copy UTF-8 to 'word_in' if(lookupwchar(bg_vowels, cc) != 0) vcount++; if((cc == '\'') || (cc == 0x300) || (cc == 0x450) || (cc == 0x45d)) { // backprime (before the vowel), combining grave accent, or accented vowel character if(cc == '\'') lex_stress = vcount+1; else lex_stress = vcount; n_stress++; if(vcount == 1) stress_first = 1; if((cc == '\'') || (cc == 0x300)) continue; // discard backprime or combining accent if(cc == 0x450) cc = 0x435; // remove accent from vowel if(cc == 0x45d) cc = 0x438; } pw += utf8_out(cc, pw); // copy UTF-8 to 'word' } *pw++ = ' '; *pw = 0; *pw1 = 0; // translate TranslateWord(translator, &word[2],0, NULL); DecodePhonemes(word_phonemes,phonemes); // find the stress position in the translation max_stress = 0; max_stress_posn = -1; vcount = 0; ph = phoneme_tab[phonPAUSE]; for(p=word_phonemes; *p != 0; p++) { ph = phoneme_tab[(unsigned int)*p]; if(ph == NULL) continue; if(ph->type == phVOWEL) vcount++; if(ph->type == phSTRESS) { if(ph->std_length > max_stress) { max_stress = ph->std_length; max_stress_posn = vcount+1; } } } if(n_stress > 1) n_stress = 1; done = 0; if(vcount < 2) { // don't list words with only one vowel } else if((lex_stress != max_stress_posn) || (n_stress != 1)) { if((vcount > 0) && (lex_stress > 0) && (lex_stress <= 7)) { if((n_stress == 2) && (stress_first)) { done = 1; fprintf(f_out,"%s\t$%d\n",&word[2],lex_stress); } if(n_stress == 1) { done = 1; fprintf(f_out,"%s\t$%d\n",&word[2],lex_stress); } } if(done == 0) { n_wrong++; fprintf(f_out,"// %s\t$text %s\n", &word[2], word_in); } if(done) n_out++; } n_words++; } fclose(f_in); fclose(f_out); CompileDictionary(path_dsource,"bg",NULL,NULL,0); delete progress; sprintf(buf,"Lexicon: Input %d, Output %d, $text %d",n_words,n_out,n_wrong); wxLogStatus(wxString(buf,wxConvLocal)); } // end of Lexicon_Bg void Lexicon_Ru() {//============== // compare stress markings in Russian RuLex file with lookup in ru_rules // Input file contains a list of Russian words (UTF-8), one per line // Stress position can be indicated either by: // A $ sign and syllable number after the word, eg: // абажура $3 // or by a + sign after the stressed vowel, eg: // абажу+ра // espeakedit produces a file: dictsource/ru_listx and a log file dictsource/ru_log int ix; char *p; int c; FILE *f_in; FILE *f_out; FILE *f_log; PHONEME_TAB *ph; int ph_code; int vcount; int ru_stress; int max_stress; int max_stress_posn; int n_words=0; int n_wrong=0; int n_errors=0; int wlength; int input_length; int check_root; char word[80]; char word2[80]; int counts[20][20][10]; char phonemes[N_WORD_PHONEMES]; char buf[200]; char fname[sizeof(path_dsource)+20]; // character codes for Russian vowels static unsigned short ru_vowels[] = {0x430,0x435,0x438,0x439,0x43e,0x443,0x44d,0x44e,0x44f,0x450,0x451,0}; typedef struct { const char *suffix; int syllables; } SUFFIX; #ifdef deleted FILE *f_roots; int sfx; const char *suffix; int wlen; int len; static SUFFIX suffixes[] = { {NULL,0}, {"ичу",2}, {"ского",2}, {"ская",2}, {"ски",1}, {"ские",2}, {"ский",1}, {"ским",1}, {"ское",2}, {"ской",1}, {"ском",1}, {"скую",2}, {"а",1}, {"е",1}, {"и",1}, {NULL,0}}; #endif memset(counts,0,sizeof(counts)); if(gui_flag) { wxString fname = wxFileSelector(_T("Read lexicon.dict"),path_dictsource, _T(""),_T(""),_T("*"),wxOPEN); if(fname.IsEmpty()) return; strcpy(buf,fname.mb_str(wxConvLocal)); } else { strcpy(buf,"lexicon.dict"); } if((f_in = fopen(buf,"r")) == NULL) { if(gui_flag) wxLogError(_T("Can't read file: ") + wxString(buf,wxConvLocal)); else fprintf(stderr,"Can't read file: %s\n",buf); return; } input_length = GetFileLength(buf); sprintf(fname,"%s%c%s",path_dsource,PATHSEP,"ru_listx"); remove(fname); // compile ru_dict without ru_listx CompileDictionary(path_dsource,"ru",NULL,NULL,0); if((f_out = fopen(fname,"w")) == NULL) { wxLogError(_T("Can't write to: ")+wxString(fname,wxConvLocal)); fclose(f_in); return; } sprintf(fname,"%s%c%s",path_dsource,PATHSEP,"ru_log"); f_log = fopen(fname,"w"); sprintf(fname,"%s%c%s",path_dsource,PATHSEP,"ru_roots_1"); // f_roots = fopen(fname,"w"); LoadVoice("ru",0); if(gui_flag) progress = new wxProgressDialog(_T("Lexicon"),_T(""),input_length); else fprintf(stderr,"Processing lexicon.dict\n"); for(;;) { if(((n_words & 0x3ff) == 0) && gui_flag) { progress->Update(ftell(f_in)); } if(fgets(buf,sizeof(buf),f_in) == NULL) break; if((p = strstr(buf,"//")) != NULL) *p = '\n'; // truncate at comment p = buf; while((*p == ' ') || (*p == '\t')) p++; if(*p == '\n') continue; // blank line ix = 0; wlength = 0; vcount = 0; ru_stress = -1; for(;;) { p += utf8_in(&c, p); if(isspace(c)) break; if(lookupwchar(ru_vowels, c)) vcount++; if(c == '+') ru_stress = vcount; else ix += utf8_out(c, &word[ix]); } word[ix] = 0; sprintf(word2," %s ",word); // surround word by spaces before calling TranslateWord() // find the marked stress position, if it has not been marked by a + after the stressed vowel if(ru_stress == -1) { while((*p == ' ') || (*p == '\t')) p++; sscanf(p,"$%d",&ru_stress); if(ru_stress == -1) { n_errors++; fprintf(f_log,"%s",buf); continue; // stress position not found } } // translate TranslateWord(translator, &word2[1],0, NULL); DecodePhonemes(word_phonemes,phonemes); // find the stress position in the translation max_stress = 0; max_stress_posn = -1; vcount = 0; check_root = 0; ph = phoneme_tab[phonPAUSE]; for(p=word_phonemes; (ph_code = *p & 0xff) != 0; p++) { ph = phoneme_tab[ph_code]; if(ph == NULL) continue; if((ph->type == phVOWEL) && (ph_code != phonSCHWA_SHORT)) vcount++; if(ph->type == phSTRESS) { if(ph->std_length > max_stress) { max_stress = ph->std_length; max_stress_posn = vcount+1; } } } n_words++; if(ru_stress > vcount) { if(f_log != NULL) { fprintf(f_log,"%s\t $%d\t // %s\n",word,ru_stress,phonemes); } n_errors++; } else { counts[vcount][ru_stress][ph->type]++; if((vcount > 1) && (ru_stress != max_stress_posn)) { n_wrong++; if((ru_stress==0) || (ru_stress > 7)) { fprintf(f_out,"// "); // we only have $1 to $7 to indicate stress position if(f_log != NULL) { fprintf(f_log,"%s\t $%d\t // %s\n",word,ru_stress,phonemes); } n_errors++; } else check_root = 1; #define X_COMPACT fprintf(f_out,"%s",word); #ifdef X_COMPACT if(wlength < 8) fputc('\t',f_out); if(wlength < 16) fputc('\t',f_out); fprintf(f_out," $%d\n",ru_stress); #else while(wlength++ < 20) fputc(' ',f_out); fprintf(f_out," $%d //%d %s\n",ru_stress,max_stress_posn,phonemes); #endif //CharStats(); } } #ifdef deleted if(check_root) { // does this word match any suffixes ? wlen = strlen(word); for(sfx=0;(suffix = suffixes[sfx].suffix) != NULL; sfx++) { len = strlen(suffix); if(len >= (wlen-2)) continue; if(ru_stress > (vcount - suffixes[sfx].syllables)) continue; if(strcmp(suffix,&word[wlen-len])==0) { strcpy(word2,word); word2[wlen-len] = 0; // fprintf(f_roots,"%s\t $%d\t\\ %s\n",word2,ru_stress,suffix); fprintf(f_roots,"%s\t $%d\n",word2,ru_stress); } } } #endif } fclose(f_in); fclose(f_out); // fclose(f_roots); sprintf(buf,"Lexicon: Total %d OK %d fixed %d errors %d (see ru_log)",n_words,n_words-n_wrong,n_wrong,n_errors); if(gui_flag) { delete progress; wxLogStatus(wxString(buf,wxConvLocal)); } else { fprintf(stderr,"%s\n",buf); } if(f_log != NULL) { #ifdef deleted // list tables of frequency of stress position for words of different syllable lengths int j,k; for(ix=2; ix<12; ix++) { fprintf(f_log,"%2d syllables\n",ix); for(k=0; k<10; k++) { fprintf(f_log," %2d :",k); for(j=1; j<=ix; j++) { fprintf(f_log,"%6d ",counts[ix][j][k]); } fprintf(f_log,"\n"); } fprintf(f_log,"\n\n"); } #endif fclose(f_log); } } // end of Lexicon_Ru void CompareLexicon(int id) {//======================== switch(id) { case MENU_LEXICON_RU: Lexicon_Ru(); break; case MENU_LEXICON_BG: Lexicon_Bg(); break; case MENU_LEXICON_DE: Lexicon_De(); break; case MENU_LEXICON_IT: Lexicon_It(1); Lexicon_It(2); break; case MENU_LEXICON_TEST: Lexicon_Test(); break; } } // end of CompareLexicon //****************************************************************************************************** extern int HashDictionary(const char *string); static int n_words; struct wcount { struct wcount *link; int count; char *word; }; static int wfreq_sorter(wcount **p1, wcount **p2) {//============================================== int x; wcount *a, *b; a = *p1; b = *p2; if((x = b->count - a->count) != 0) return(x); return(strcmp(a->word,b->word)); } static void wfreq_add(const char *word, wcount **hashtab) {//====================================================== wcount *p; wcount **p2; int len; int hash; hash = HashDictionary(word); p2 = &hashtab[hash]; p = *p2; while(p != NULL) { if(strcmp(p->word,word)==0) { p->count++; return; } p2 = &p->link; p = *p2; } // word not found, add it to the list len = strlen(word) + 1; if((p = (wcount *)malloc(sizeof(wcount)+len)) == NULL) return; p->count = 1; p->link = NULL; p->word = (char *)p + sizeof(wcount); strcpy(p->word,word); *p2 = p; n_words++; } void CountWordFreq(wxString path, wcount **hashtab) {//================================================ // Count the occurances of words in this file FILE *f_in; unsigned char c; int wc; unsigned int ix, j, k; int n_chars; char buf[80]; char wbuf[80]; if((f_in = fopen(path.mb_str(wxConvLocal),"rb")) == NULL) return; while(!feof(f_in)) { while((c = fgetc(f_in)) < 'A') { // skip leading spaces, numbers, etc if(feof(f_in)) break; } // read utf8 bytes until a space, number or punctuation ix = 0; while(!feof(f_in) && (c >= 'A') && (ix < sizeof(buf)-1)) { buf[ix++] = c; c = fgetc(f_in); } buf[ix++] = 0; buf[ix] = 0; // the buf may contain non-alphabetic characters j = 0; n_chars = 0; for(k=0; k 2) { wfreq_add(wbuf,hashtab); } j = 0; n_chars = 0; } } } fclose(f_in); } // end of CountWordFreq void MakeWordFreqList() {//==================== // Read text files from a specified directory and make a list of the most frequently occuring words. struct wcount *whashtab[N_HASH_DICT]; wcount **w_list; int ix; int j; int hash; wcount *p; FILE *f_out; char buf[200]; char buf2[200]; wxString dir = wxDirSelector(_T("Directory of text files"),path_speaktext); if(dir.IsEmpty()) return; memset(whashtab,0,sizeof(whashtab)); wxString path = wxFindFirstFile(dir+_T("/*"),wxFILE); while (!path.empty()) { if(path.AfterLast(PATHSEP) != _T("!wordcounts")) { CountWordFreq(path,whashtab); path = wxFindNextFile(); } } // put all the words into a list and then sort it w_list = (wcount **)malloc(sizeof(wcount *) * n_words); ix = 0; for(hash=0; hash < N_HASH_DICT; hash++) { p = whashtab[hash]; while((p != NULL) && (ix < n_words)) { w_list[ix++] = p; p = p->link; } } qsort((void *)w_list,ix,sizeof(wcount *),(int(*)(const void *,const void *))wfreq_sorter); // write out the sorted list strcpy(buf,dir.mb_str(wxConvLocal)); sprintf(buf2,"%s/!wordcounts",buf); if((f_out = fopen(buf2,"w")) == NULL) return; for(j=0; jcount,p->word); free(p); } fclose(f_out); } // end of Make WorkFreqList //****************************************************************************************************** void ConvertToUtf8() {//================= // Convert a file from 8bit to UTF8, according to the current voice unsigned int c; int ix; FILE *f_in; FILE *f_out; char buf[200]; wxString fname = wxFileSelector(_T("Convert file to UTF8"),wxString(path_home,wxConvLocal), _T(""),_T(""),_T("*"),wxOPEN); if(fname.IsEmpty()) return; strcpy(buf,fname.mb_str(wxConvLocal)); f_in = fopen(buf,"r"); if(f_in == NULL) { wxLogError(_T("Can't read file: ")+fname); return; } strcat(buf,"_1"); f_out = fopen(buf,"w"); if(f_out == NULL) { wxLogError(_T("Can't create file: ")+wxString(buf,wxConvLocal)); fclose(f_in); return; } while(!feof(f_in)) { c = fgetc(f_in); if(c >= 0xa0) c = translator->charset_a0[c-0xa0]; ix = utf8_out(c,buf); fwrite(buf,ix,1,f_out); } fclose(f_in); fclose(f_out); wxLogStatus(_T("Written to: ")+fname+_T("_1")); } // end of ConvertToItf8 //****************************************************************************************************** #define N_SORT_LIST 10000 void DictionarySort(const char *dictname) {//====================================== // Sort rules in *_rules file between lines which begin with //sort and //endsort FILE *f_in; FILE *f_out; int ix; char *p; char *p_end; char *p_pre; int sorting; int sort_ix=0; int sort_count=0; int line_len; int key_len; char buf[200]; char key[200]; char fname_in[200]; char fname_out[200]; char *sort_list[N_SORT_LIST]; wxLogMessage(_T("Sorts the *_rules file, between lines which begin with\n//sort\n and\n//endsort")); // try with and without '.txt' extension sprintf(fname_in,"%s%s_rules.txt",path_dsource,dictname); if((f_in = fopen(fname_in,"r")) == NULL) { sprintf(fname_in,"%s%s_rules",path_dsource,dictname); if((f_in = fopen(fname_in,"r")) == NULL) { wxLogError(_T("Can't open rules file: ") + wxString(fname_in,wxConvLocal)); return; } } sprintf(fname_out,"%s%s_rules_sorted",path_dsource,dictname); if((f_out = fopen(fname_out,"w")) == NULL) { wxLogError(_T("Can't write to file: ") + wxString(fname_out,wxConvLocal)); fclose(f_in); return; } sorting = 0; while(fgets(buf, sizeof(buf)-1, f_in) != NULL) { buf[sizeof(buf)-1] = 0; // ensure zero byte terminator line_len = strlen(buf); if(memcmp(buf,"//endsort",9)==0) { sort_count++; sorting = 0; qsort((void *)sort_list, sort_ix, sizeof(char *), (int(*)(const void *, const void *))string_sorter); // write out the sorted lines for(ix=0; ix= N_SORT_LIST) { wxLogError(_T("Too many lines to sort, > %d"), N_SORT_LIST); break; } } fclose(f_in); fclose(f_out); if(sorting != 0) { wxLogError(_T("Missing //$endsort")); } wxLogStatus(_T("Sorted %d sections. Written to file: ") + wxString(fname_out,wxConvLocal),sort_count); } // end of DictionarySort void DictionaryFormat(const char *dictname) {//======================================== // Format the *_rules file for the current voice FILE *f_in; FILE *f_out; char *p; char *p_start; unsigned short *pw; unsigned short *pw_match; unsigned short *pw_post = NULL; unsigned short *pw_phonemes = NULL; int c; int ix; int n_pre; int n_match; int n_post; int n_phonemes; int n_spaces; int n_out; int formatting; int comment; char buf[200]; unsigned short bufw[200]; char conditional[80]; char fname_in[200]; char fname_out[200]; const int tab1 = 8; const int tab2 = 18; const int tab3 = 28; // try with and without '.txt' extension sprintf(fname_in,"%s%s_rules.txt",path_dsource,dictname); if((f_in = fopen(fname_in,"r")) == NULL) { sprintf(fname_in,"%s%s_rules",path_dsource,dictname); if((f_in = fopen(fname_in,"r")) == NULL) { wxLogError(_T("Can't open rules file: ") + wxString(fname_in,wxConvLocal)); return; } } sprintf(fname_out,"%s%s_rules_formatted",path_dsource,dictname); if((f_out = fopen(fname_out,"w")) == NULL) { wxLogError(_T("Can't write to file: ") + wxString(fname_out,wxConvLocal)); fclose(f_in); return; } formatting = 0; n_match = 0; while(fgets(buf, sizeof(buf)-1, f_in) != NULL) { buf[sizeof(buf)-1] = 0; // ensure zero byte terminator ix = strlen(buf) - 1; while((buf[ix]=='\n') || (buf[ix]==' ') || (buf[ix]=='\t')) ix--; buf[ix+1] = 0; // strip tailing spaces p_start = buf; while((*p_start==' ') || (*p_start == '\t')) p_start++; // skip leading spaces comment = 0; if((p_start[0]=='/') && (p_start[1]=='/')) comment = 1; ix = 0; if(*p_start == '?') { // conditional rule while(!isspace(*p_start) && (*p_start != 0)) { conditional[ix++] = *p_start++; } while((*p_start == ' ') || (*p_start == '\t')) p_start++; } conditional[ix] = 0; if(buf[0] == '.') { formatting = 0; } if(memcmp(p_start, ".group", 6) == 0) { formatting = 2; if(n_match > 0) { // previous line was not blank, so add a blank line fprintf(f_out,"\n"); } } n_match = 0; if((formatting == 1) && (comment==0)) { // convert from UTF-8 to UTF-16 p = p_start; pw = bufw; do { p += utf8_in(&c, p); *pw++ = c; } while (c != 0); pw = bufw; while((*pw != ')') && (*pw != 0) && !iswspace(*pw)) pw++; n_pre = 0; n_post = 0; n_phonemes = 0; n_spaces = 0; if(*pw != 0) n_spaces = tab1; if(*pw == ')') { // there is a pre-condition n_pre = pw - bufw + 1; n_spaces = tab1 - n_pre - 1; pw++; while((*pw==' ') || (*pw=='\t')) pw++; } else { pw = bufw; } pw_match = pw; while(((c = *pw)!= ' ') && (c != '\t') && (c != '(') && (c != 0)) { pw++; } n_match = pw - pw_match; while(((c = *pw)==' ') || (c == '\t')) pw++; if(*pw == '(') { pw_post = pw; while(((c = *pw)!=' ') && (c != '\t') && (c != 0)) pw++; n_post = pw - pw_post; while(((c = *pw)==' ') || (c == '\t')) pw++; } if((*pw != 0) && ((*pw != '/') || (pw[1] != '/'))) { pw_phonemes = pw; while(((c = *pw)!=' ') && (c != '\t') && (c != 0)) pw++; n_phonemes = pw - pw_phonemes; while(((c = *pw)==' ') || (c == '\t')) pw++; } // write formatted line p = buf; if(conditional[0] != 0) { ix = 0; while(conditional[ix] != 0) { *p++ = conditional[ix++]; n_spaces--; } *p++ = ' '; n_spaces--; } while(n_spaces-- > 0) { *p++ = ' '; } if(n_pre > 0) { ix = 0; for(ix=0; ix 0) { ix = 0; for(ix=0; ix 0) { for(ix=0; ix= tab1) n_out += (n_pre - tab1 + 1); if(n_phonemes > 0) { n_spaces = tab2 - n_out; while(n_spaces-- > 0) { *p++ = ' '; n_out++; } // write the phoneme codes for(ix=0; ix 0) { *p++ = ' '; } } // write the remainer of the line while(*pw != 0) { p+= utf8_out(*pw++, p); } *p = 0; } if(formatting > 1) formatting--; fprintf(f_out, "%s\n", buf); } fclose(f_in); fclose(f_out); remove(fname_in); if(rename(fname_out, fname_in) == 0) wxLogStatus(_("Written to file: ") + wxString(fname_in,wxConvLocal)); else wxLogStatus(_("Failed to rename: ") + wxString(fname_out,wxConvLocal)); } // end of DictionaryFormat //****************************************************************************************************** //#define calcspeedtab #ifdef calcspeedtab // used to set up the presets in the speed_lookup table // interpolate between a set of measured wpm values void SetSpeedTab(void) {//=================== #define N_WPM 19 // Interpolation table to translate from words-per-minute to internal speed // words-per-minute values (measured) static float wpm1[N_WPM] = {0, 82, 96, 108, 124, 134, 147, 162, 174, 189, 224, 259, 273, 289, 307, 326, 346, 361, 370 }; // corresponding internal speed values static float wpm2[N_WPM] = {0,253,200, 170, 140, 125, 110, 95, 85, 75, 55, 40, 35, 30, 25, 20, 15, 10, 5 }; unsigned char speed_lookup[290]; unsigned int ix; float x; int speed_wpm; FILE *f; // convert from word-per-minute to internal speed code for(speed_wpm=80; speed_wpm<370; speed_wpm++) { for(ix=2; ix lc else { fprintf(f,"%4d,",y); } if((ix&15)==15) fprintf(f," // %x\n ",ix & ~15); } fprintf(f,"\n%s\n ",charset); for(ix=0x80; ix<0x100; ix++) { inbuf[0] = ix; inbuf[1] = 0; inbuf[2] = 0; outbuf[0] = 0; outbuf[1] = 0; n_inbuf = 2; n_outbuf = sizeof(outbuf); p_inbuf = (char *)inbuf; p_outbuf = (char *)outbuf; n = iconv(cd, &p_inbuf, &n_inbuf, &p_outbuf, &n_outbuf); fprintf(f,"0x%.2x%.2x, ",outbuf[1],outbuf[0]); if((ix&7)==7) fprintf(f,"// %.2x\n ",ix & ~7); } fclose(f); iconv_close(cd); } #endif #ifdef deleted void Test2() { // char buf[120]; FILE *f; FILE *f_out; unsigned char *p; f = fopen("/home/jsd1/tmp1/list","r"); if(f == NULL) return; f_out = fopen("/home/jsd1/tmp1/list_out","w"); if(f_out == NULL) return; while(!feof(f)) { if(fgets(buf,sizeof(buf),f) == NULL) break; p = (unsigned char *)buf; while(*p > ' ') p++; *p = 0; fprintf(f_out,"%s . . .\n",buf); } fclose(f); fclose(f_out); } #endif const char* text1 = "Hello world. Testing."; extern void TestCompile2(void); void TestTest(int control) {//======================= FILE *f; unsigned int c; unsigned int ix=0; char textbuf[2000]; espeak_VOICE voice; static unsigned int unique_identifier= 123; static int user_data = 456; //CharsetToUnicode("ISO-8859-4"); //CharsetToUnicode("ISCII"); if(control==2) { return; } memset(&voice,0,sizeof(voice)); f = fopen("/home/jsd1/speechdata/text/test.txt","r"); if(f==NULL) return; while(!feof(f) && (ix < sizeof(textbuf)-2)) { c = fgetc(f); if(!feof(f)) textbuf[ix++] = c; } textbuf[ix] = 0; fclose(f); f_wavtest = OpenWaveFile3("/home/jsd1/speechdata/text/test.wav"); f_events = fopen("/home/jsd1/speechdata/text/events","w"); fprintf(f_events,"Type Audio Text Length Id\n"); fclose(f_events); espeak_Initialize(AUDIO_OUTPUT_RETRIEVAL, 1000, NULL, espeakINITIALIZE_PHONEME_IPA); espeak_SetSynthCallback(TestSynthCallback); espeak_SetUriCallback(TestUriCallback); espeak_SetVoiceByName("en"); espeak_Synth(textbuf, strlen(textbuf)+1, 0, POS_CHARACTER, 0, espeakSSML|espeakCHARS_UTF8, &unique_identifier, (void *)user_data); // espeak_Synth(text1, strlen(text1)+1, 0, POS_CHARACTER, 0, espeakSSML|espeakCHARS_UTF8, &unique_identifier, (void *)(user_data+1)); espeak_SetParameter(espeakPUNCTUATION, 1, 0); espeak_Synchronize(); // espeak_Cancel(); espeak_SetParameter(espeakPUNCTUATION, 1, 0); }