[1.30.04] SSML <audio src=""> now works without using a UriCallback() function, and works with the command-line espeak.

Pitch changes for mbrola are now applied over the first 80% of the phoneme, not over the whole length of the phoneme.  This seems to give better intonation for mbrola voices.


git-svn-id: https://espeak.svn.sourceforge.net/svnroot/espeak/trunk@118 d46cf337-b52f-0410-862d-fd96e6ae7743

src/compiledata.cpp

@@ -73,6 +73,7 @@
 #define tINCLUDE  31
 
 extern void Write4Bytes(FILE *f, int value);
+extern int Read4Bytes(FILE *f);
 extern void MakeVowelLists(void);
 extern void FindPhonemesUsed(void);
 extern void DrawEnvelopes();
@@ -309,21 +310,6 @@ static unsigned int StringToWord(const char *string)
 }
 
 
-int Read4Bytes(FILE *f)
-{//====================
-// Read 4 bytes (least significant first) into a word
-	int ix;
-	unsigned char c;
-	int acc=0;
-
-	for(ix=0; ix<4; ix++)
-	{
-		c = fgetc(f) & 0xff;
-		acc += (c << (ix*8));
-	}
-	return(acc);
-}
-
 
 static FILE *fopen_log(FILE *f_log, const char *fname,const char *access)
 {//==================================================

src/readclause.cpp

@@ -511,6 +511,21 @@ const char *Translator::LookupCharName(int c)
 	return(buf);
 }
 
+int Read4Bytes(FILE *f)
+{//====================
+// Read 4 bytes (least significant first) into a word
+	int ix;
+	unsigned char c;
+	int acc=0;
+
+	for(ix=0; ix<4; ix++)
+	{
+		c = fgetc(f) & 0xff;
+		acc += (c << (ix*8));
+	}
+	return(acc);
+}
+
 
 static int LoadSoundFile(const char *fname, int index)
 {//===================================================
@@ -520,7 +535,7 @@ static int LoadSoundFile(const char *fname, int index)
 	int  length;
 	char fname_temp[100];
 	char fname2[sizeof(path_home)+13+40];
-	char command[sizeof(fname2)+sizeof(fname2)+30];
+	char command[sizeof(fname2)+sizeof(fname2)+40];
 
 	if(fname == NULL)
 	{
@@ -542,16 +557,27 @@ static int LoadSoundFile(const char *fname, int index)
 #ifdef PLATFORM_POSIX
 	if((f = fopen(fname,"rb")) != NULL)
 	{
-		int header[11];
-		fread(header, 1, 44, f);
+		int ix;
+		const char *resample;
+		int header[3];
 
-		if(header[6] != samplerate)
+		fseek(f,20,SEEK_SET);
+		for(ix=0; ix<3; ix++)
+			header[ix] = Read4Bytes(f);
+
+		// if the sound file is not mono, 16 bit signed, at the correct sample rate, then convert it
+		if((header[0] != 0x10001) || (header[1] != samplerate) || (header[2] != samplerate*2))
 		{
 			fclose(f);
 			f = NULL;
 
+			if(header[2] == samplerate)
+				resample = "";
+			else
+				resample = "polyphase";
+
 			sprintf(fname_temp,"%s.wav",tmpnam(NULL));
-			sprintf(command,"sox \"%s\" -r %d -w %s polyphase\n",fname,samplerate,fname_temp);
+			sprintf(command,"sox \"%s\" -r %d -w -s -c1 %s %s\n", fname, samplerate, fname_temp, resample);
 			if(system(command) == 0)
 			{
 				fname = fname_temp;
@@ -572,14 +598,18 @@ static int LoadSoundFile(const char *fname, int index)
 
 	length = GetFileLength(fname);
 	fseek(f,0,SEEK_SET);
-	p = Alloc(length);
+	if((p = (char *)realloc(soundicon_tab[index].data, length)) == NULL)
+	{
+		fclose(f);
+		return(4);
+	}
 	fread(p,length,1,f);
 	fclose(f);
 	remove(fname_temp);
 
 	ip = (int *)(&p[40]);
 	soundicon_tab[index].length = (*ip) / 2;  // length in samples
-	soundicon_tab[index].data = p+44;  // skip WAV header
+	soundicon_tab[index].data = p;
 	return(0);
 }  //  end of LoadSoundFile
 
@@ -589,7 +619,7 @@ static int LookupSoundicon(int c)
 // Find the sound icon number for a punctuation chatacter
 	int ix;
 
-	for(ix=0; ix<n_soundicon_tab; ix++)
+	for(ix=N_SOUNDICON_SLOTS; ix<n_soundicon_tab; ix++)
 	{
 		if(soundicon_tab[ix].name == c)
 		{
@@ -605,6 +635,34 @@ static int LookupSoundicon(int c)
 }
 
 
+static int LoadSoundFile2(const char *fname)
+{//=========================================
+// Load a sound file into one of the reserved slots in the sound icon table
+// (if it'snot already loaded)
+
+	int ix;
+	static int slot = -1;
+
+	for(ix=0; ix<n_soundicon_tab; ix++)
+	{
+		if(((soundicon_tab[ix].filename != NULL) && strcmp(fname, soundicon_tab[ix].filename) == 0))
+			return(ix);   // already loaded
+	}
+
+	// load the file into the next slot
+	slot++;
+	if(slot >= N_SOUNDICON_SLOTS)
+		slot = 0;
+
+	if(LoadSoundFile(fname, slot) != 0)
+		return(-1);
+
+	soundicon_tab[slot].filename = (char *)realloc(soundicon_tab[ix].filename, strlen(fname)+1);
+	strcpy(soundicon_tab[slot].filename, fname);
+	return(slot);
+}
+
+
 
 int Translator::AnnouncePunctuation(int c1, int c2, char *buf, int bufix)
 {//======================================================================
@@ -1462,12 +1520,27 @@ static int ProcessSsmlTag(wchar_t *xml_buf, char *outbuf, int &outix, int n_outb
 
 		if((attr1 = GetSsmlAttribute(px,"src")) != NULL)
 		{
+			char fname[256];
 			attrcopy_utf8(buf,attr1,sizeof(buf));
 
 			if(uri_callback == NULL)
 			{
-// we could load the audio file as a sound icon (assuming it's a local WAV file)
-// But when to free() the data?
+				if((xmlbase != NULL) && (buf[0] != '/'))
+				{
+					sprintf(fname,"%s/%s",xmlbase,buf);
+					index = LoadSoundFile2(fname);
+				}
+				else
+				{
+					index = LoadSoundFile2(buf);
+				}
+				if(index >= 0)
+				{
+					sprintf(buf,"%c%dI",CTRL_EMBEDDED,index);
+					strcpy(&outbuf[outix],buf);
+					outix += strlen(buf);
+					sp->parameter[espeakSILENCE] = 1;
+				}
 			}
 			else
 			{
@@ -1853,6 +1926,25 @@ f_input = f_in;  // for GetC etc
 
 		if(iswalnum(c1))
 			any_alnum = 1;
+		else
+		if(iswspace(c1))
+		{
+			if(translator_name == 0x6a626f)
+			{
+				// language jbo : lojban
+				// treat "i" or ".i" as end-of-sentence
+				if(buf[ix-1] == 'i')
+				{
+					if(iswspace(buf[ix-2]) || ((buf[ix-2] == '.') && iswspace(buf[ix-3])))
+					{
+						UngetC(c2);
+						buf[ix] = ' ';
+						buf[ix+1] = 0;
+						return(CLAUSE_PERIOD);
+					}
+				}
+			}
+		}
 
 		if(iswupper(c1))
 		{

src/speak.cpp

@@ -305,9 +305,7 @@ static int initialise(void)
 		else
 			fprintf(stderr,"Wrong version of espeak-data 0x%x (expects 0x%x)\n",result,version_phdata);
 	}
-#ifndef PLATFORM_WINDOWS
-	LoadConfig();  // causes problem on Windows, don't know why
-#endif
+	LoadConfig();
 	SetVoiceStack(NULL);
 	SynthesizeInit();
 

src/speak_lib.cpp

@@ -335,9 +335,7 @@ static int initialise(void)
 	int param;
 	int result;
 
-#ifndef __WIN32__
 	LoadConfig();
-#endif
 	WavegenInit(22050,0);   // 22050
 	if((result = LoadPhData()) != 1)
 	{

src/synth_mbrola.cpp

@@ -252,8 +252,9 @@ int GetMbrName(PHONEME_LIST *plist, PHONEME_TAB *ph, PHONEME_TAB *ph_prev, PHONE
 }
 
 
-static char *WritePitch(int env, int pitch1, int pitch2, int split)
-{//================================================================
+static char *WritePitch(int env, int pitch1, int pitch2, int split, int final)
+{//===========================================================================
+// final=1:  only give the final pitch value.
 	int x;
 	int ix;
 	int pitch_base;
@@ -264,11 +265,12 @@ static char *WritePitch(int env, int pitch1, int pitch2, int split)
 	int min = 999;
 	int y_max=0;
 	int y_min=0;
+	int env100 = 80;  // apply the pitch change only over this proportion of the mbrola phoneme(s)
 	int y2;
 	int y[4];
 	int env_split;
-	char buf[40];
-	static char output[40];
+	char buf[50];
+	static char output[50];
 
 	output[0] = 0;
 	pitch_env = envelope_data[env];
@@ -335,18 +337,18 @@ static char *WritePitch(int env, int pitch1, int pitch2, int split)
 
 			if(split > 0)
 			{
-				y2 = (y[ix] * 100)/env_split;
+				y2 = (y[ix] * env100)/env_split;
 			}
 			else
 			if(split < 0)
 			{
-				y2 = ((y[ix]-env_split) * 100)/env_split;
+				y2 = ((y[ix]-env_split) * env100)/env_split;
 			}
 			else
 			{
-				y2 = (y[ix] * 100)/128;
+				y2 = (y[ix] * env100)/128;
 			}
-			if((y2 > 0) && (y2 <= 100))
+			if((y2 > 0) && (y2 <= env100))
 			{
 				sprintf(buf," %d %d",y2,p2/4096);
 				strcat(output,buf);
@@ -354,14 +356,23 @@ static char *WritePitch(int env, int pitch1, int pitch2, int split)
 		}
 	}
 
+	p_end = p_end/4096;
 	if(split <= 0)
 	{
-		sprintf(buf," 100 %d",p_end/4096);
+		sprintf(buf," %d %d",env100,p_end);
+		strcat(output,buf);
+	}
+	if(env100 < 100)
+	{
+		sprintf(buf," %d %d",100,p_end);
 		strcat(output,buf);
 	}
 	strcat(output,"\n");
+
+	if(final)
+		sprintf(output,"\t100 %d\n",p_end);
 	return(output);
-}  // end of SetPitch
+}  // end of WritePitch
 
 
 #ifdef USE_MBROLA_LIB
@@ -506,6 +517,7 @@ void MbrolaTranslate(PHONEME_LIST *plist, int n_phonemes, FILE *f_mbrola)
 	int control;
 	int done;
 	int len_percent;
+	const char *final_pitch;
 	char buf[80];
 	char mbr_buf[120];
 
@@ -562,7 +574,7 @@ void MbrolaTranslate(PHONEME_LIST *plist, int n_phonemes, FILE *f_mbrola)
 				len = 1;
 		}
 		else
-			len = (70 * speed_factor2)/256;
+			len = (80 * speed_factor2)/256;
 
 		sprintf(buf,"%s\t",WordToString(name));
 		strcat(mbr_buf,buf);
@@ -575,6 +587,7 @@ void MbrolaTranslate(PHONEME_LIST *plist, int n_phonemes, FILE *f_mbrola)
 		}
 
 		done = 0;
+		final_pitch = "";
 
 		switch(ph->type)
 		{
@@ -589,16 +602,16 @@ void MbrolaTranslate(PHONEME_LIST *plist, int n_phonemes, FILE *f_mbrola)
 
 			if(name2 == 0)
 			{
-				sprintf(buf,"%d\t%s", len, WritePitch(p->env,p->pitch1,p->pitch2,0));
+				sprintf(buf,"%d\t%s", len, WritePitch(p->env,p->pitch1,p->pitch2,0,0));
 				strcat(mbr_buf,buf);
 			}
 			else
 			{
 				len1 = (len * len_percent)/100;
-				sprintf(buf,"%d\t%s", len1, WritePitch(p->env,p->pitch1,p->pitch2,len_percent));
+				sprintf(buf,"%d\t%s", len1, WritePitch(p->env,p->pitch1,p->pitch2,len_percent,0));
 				strcat(mbr_buf,buf);
 
-				sprintf(buf,"%s\t%d\t%s", WordToString(name2), len-len1, WritePitch(p->env,p->pitch1,p->pitch2,-len_percent));
+				sprintf(buf,"%s\t%d\t%s", WordToString(name2), len-len1, WritePitch(p->env,p->pitch1,p->pitch2,-len_percent,0));
 				strcat(mbr_buf,buf);
 			}
 			done = 1;
@@ -635,6 +648,17 @@ void MbrolaTranslate(PHONEME_LIST *plist, int n_phonemes, FILE *f_mbrola)
 			{
 				len = DoSpect(p->ph,prev->ph,phoneme_tab[phonPAUSE],2,p,-1);
 				len = (len * 1000)/samplerate;
+				if(next->type == phPAUSE)
+					len += 50;
+				final_pitch = WritePitch(p->env,p->pitch1,p->pitch2,0,1);
+			}
+			break;
+
+		case phLIQUID:
+			if(next->type == phPAUSE)
+			{
+				len += 50;
+				final_pitch = WritePitch(p->env,p->pitch1,p->pitch2,0,1);
 			}
 			break;
 		}
@@ -648,7 +672,7 @@ void MbrolaTranslate(PHONEME_LIST *plist, int n_phonemes, FILE *f_mbrola)
 				strcat(mbr_buf,buf);
 				len -= len1;
 			}
-			sprintf(buf,"%d\n",len);
+			sprintf(buf,"%d%s\n",len,final_pitch);
 			strcat(mbr_buf,buf);
 		}
 

src/synthdata.cpp

@@ -35,7 +35,7 @@
 #include "translate.h"
 #include "wave.h"
 
-const char *version_string = "1.30.03  04.Jan.08";
+const char *version_string = "1.30.04  07.Jan.08";
 const int version_phdata  = 0x013000;
 
 int option_device_number = -1;
@@ -622,6 +622,12 @@ void LoadConfig(void)
 	char *p;
 	char string[200];
 
+	for(ix=0; ix<N_SOUNDICON_SLOTS; ix++)
+	{
+		soundicon_tab[ix].filename = NULL;
+		soundicon_tab[ix].data = NULL;
+	}
+
 	sprintf(buf,"%s%c%s",path_home,PATHSEP,"config");
 	if((f = fopen(buf,"r"))==NULL)
 	{

src/synthesize.cpp

@@ -62,7 +62,7 @@ static int  syllable_centre;
 
 static voice_t *new_voice=NULL;
 
-int n_soundicon_tab=0;
+int n_soundicon_tab=N_SOUNDICON_SLOTS;
 SOUND_ICON soundicon_tab[N_SOUNDICON_TAB];
 
 #define RMS_GLOTTAL1 35   // vowel before glottal stop
@@ -1041,13 +1041,13 @@ static void DoEmbedded(int &embix, int sourceix)
 		case EMBED_I:   // play dynamically loaded wav data (sound icon)
 			if((int)value < n_soundicon_tab)
 			{
-				if((wcmdq[wcmdq_tail][1] = soundicon_tab[value].length) != 0)
+				if(soundicon_tab[value].length != 0)
 				{
 					DoPause(10);   // ensure a break in the speech
 					wcmdq[wcmdq_tail][0] = WCMD_WAVE;
 					wcmdq[wcmdq_tail][1] = soundicon_tab[value].length;
-					wcmdq[wcmdq_tail][2] = (long)soundicon_tab[value].data;
-					wcmdq[wcmdq_tail][3] = 0x1f00;   // 16 bit data, amp=32
+					wcmdq[wcmdq_tail][2] = (long)soundicon_tab[value].data + 44;  // skip WAV header
+					wcmdq[wcmdq_tail][3] = 0x1500;   // 16 bit data, amp=21
 					WcmdqInc();
 				}
 			}

src/synthesize.h

@@ -262,7 +262,8 @@ extern int option_log_frames;
 extern const char *version_string;
 extern const int version_phdata;
 
-#define N_SOUNDICON_TAB  100
+#define N_SOUNDICON_TAB  80   // total entries in soundicon_tab
+#define N_SOUNDICON_SLOTS 4    // number of slots reserved for dynamic loading of autio files
 extern int n_soundicon_tab;
 extern SOUND_ICON soundicon_tab[N_SOUNDICON_TAB];
 

src/translate.cpp

@@ -1724,7 +1724,7 @@ void *Translator::TranslateClause(FILE *f_text, const void *vp_input, int *tone_
 	WORD_TAB words[N_CLAUSE_WORDS];
 	int word_count=0;      // index into words
 
-	char sbuf[512];
+	char sbuf[N_TR_SOURCE];
 
 	int terminator;
 	int tone;