Merge pull request #813

src/libespeak-ng/readclause.c

 	memset(p, ' ', utf8_in(&c, p));
 }
 
-static MNEM_TAB xml_char_mnemonics[] = {
-	{ "gt",   '>' },
-	{ "lt",   0xe000 + '<' },   // private usage area, to avoid confusion with XML tag
-	{ "amp",  '&' },
-	{ "quot", '"' },
-	{ "nbsp", ' ' },
-	{ "apos", '\'' },
-	{ NULL,   -1 }
-};
-
 int ReadClause(Translator *tr, char *buf, short *charix, int *charix_top, int n_buf, int *tone_type, char *voice_change)
 {
 	/* Find the end of the current clause.
 	int phoneme_mode = 0;
 	int n_xml_buf;
 	int terminator;
-	int found;
 	bool any_alnum = false;
-	bool self_closing;
 	int punct_data = 0;
 	bool is_end_clause;
 	int announced_punctuation = 0;
 				c2 = GetC();
 				sprintf(ungot_string, "%s%c%c", &xml_buf2[0], c1, c2);
 
+				int found = -1;
 				if (c1 == ';') {
-					if (xml_buf2[0] == '#') {
-						// character code number
-						if (xml_buf2[1] == 'x')
-							found = sscanf(&xml_buf2[2], "%x", (unsigned int *)(&c1));
-						else
-							found = sscanf(&xml_buf2[1], "%d", &c1);
-					} else {
-						if ((found = LookupMnem(xml_char_mnemonics, xml_buf2)) != -1) {
-							c1 = found;
-							if (c2 == 0)
-								c2 = ' ';
-						}
-					}
-				} else
-					found = -1;
+					found = ParseSsmlReference(xml_buf2, &c1, &c2);
+				}
 
 				if (found <= 0) {
 					ungot_string_ix = 0;
 				if ((c1 <= 0x20) && ((sayas_mode == SAYAS_SINGLE_CHARS) || (sayas_mode == SAYAS_KEY)))
 					c1 += 0xe000; // move into unicode private usage area
 			} else if ((c1 == '<') && (ssml_ignore_l_angle != '<')) {
-				if ((c2 == '!') || (c2 == '?')) {
-					// a comment, ignore until closing '<'  (or <?xml tag )
-					while (!Eof() && (c1 != '>'))
-						c1 = GetC();
-					c2 = ' ';
-				} else if ((c2 == '/') || iswalpha(c2)) {
+				if ((c2 == '/') || iswalpha(c2) || c2 == '!' || c2 == '?') {
 					// check for space in the output buffer for embedded commands produced by the SSML tag
 					if (ix > (n_buf - 20)) {
 						// Perhaps not enough room, end the clause before the SSML tag
 					xml_buf[n_xml_buf] = 0;
 					c2 = ' ';
 
-					self_closing = false;
-					if (xml_buf[n_xml_buf-1] == '/') {
-						// a self-closing tag
-						xml_buf[n_xml_buf-1] = ' ';
-						self_closing = true;
-					}
-
-					terminator = ProcessSsmlTag(xml_buf, buf, &ix, n_buf, self_closing, xmlbase, &audio_text, current_voice_id, &base_voice, base_voice_variant_name, &ignore_text, &clear_skipping_text, &sayas_mode, &sayas_start, ssml_stack, &n_ssml_stack, &n_param_stack, (int *)speech_parameters);
+					terminator = ProcessSsmlTag(xml_buf, buf, &ix, n_buf, xmlbase, &audio_text, current_voice_id, &base_voice, base_voice_variant_name, &ignore_text, &clear_skipping_text, &sayas_mode, &sayas_start, ssml_stack, &n_ssml_stack, &n_param_stack, (int *)speech_parameters);
 
 					if (terminator != 0) {
 						buf[ix] = ' ';

src/libespeak-ng/ssml.c

 	}
 }
 
-int ProcessSsmlTag(wchar_t *xml_buf, char *outbuf, int *outix, int n_outbuf, bool self_closing, const char *xmlbase, bool *audio_text, char *current_voice_id, espeak_VOICE *base_voice, char *base_voice_variant_name, bool *ignore_text, bool *clear_skipping_text, int *sayas_mode, int *sayas_start, SSML_STACK *ssml_stack, int *n_ssml_stack, int *n_param_stack, int *speech_parameters)
+int ProcessSsmlTag(wchar_t *xml_buf, char *outbuf, int *outix, int n_outbuf, const char *xmlbase, bool *audio_text, char *current_voice_id, espeak_VOICE *base_voice, char *base_voice_variant_name, bool *ignore_text, bool *clear_skipping_text, int *sayas_mode, int *sayas_start, SSML_STACK *ssml_stack, int *n_ssml_stack, int *n_param_stack, int *speech_parameters)
 {
 	// xml_buf is the tag and attributes with a zero terminator in place of the original '>'
 	// returns a clause terminator value.
 	PARAM_STACK *sp;
 	SSML_STACK *ssml_sp;
 
+	// don't process comments and xml declarations
+	if (wcsncmp(xml_buf, (wchar_t *) "!--", 3) == 0 || wcsncmp(xml_buf, (wchar_t *) "?xml", 4) == 0) {
+		return 0;
+		}
+
 	// these tags have no effect if they are self-closing, eg. <voice />
 	static char ignore_if_self_closing[] = { 0, 1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 0, 1, 0, 1, 0, 0, 0, 0 };
 
+	bool self_closing = false;
+	int len;
+	len = wcslen(xml_buf);
+	if (xml_buf[len - 1] == '/') {
+		// a self-closing tag
+		xml_buf[len - 1] = ' ';
+		self_closing = true;
+	}
+
 	static const MNEM_TAB mnem_phoneme_alphabet[] = {
 		{ "espeak", 1 },
 		{ NULL,    -1 }
 	}
 	return 0;
 }
+
+static MNEM_TAB xml_entity_mnemonics[] = {
+	{ "gt",   '>' },
+	{ "lt",   0xe000 + '<' },   // private usage area, to avoid confusion with XML tag
+	{ "amp",  '&' },
+	{ "quot", '"' },
+	{ "nbsp", ' ' },
+	{ "apos", '\'' },
+	{ NULL,   -1 }
+};
+
+int ParseSsmlReference(char *ref, int *c1, int *c2) {
+	// Check if buffer *ref contains an XML character or entity reference
+	// if found, set *c1 to the replacement char
+	// change *c2 for entity references
+	// returns >= 0 on success
+
+	if (ref[0] == '#') {
+		// character reference
+		if (ref[1] == 'x')
+			return sscanf(&ref[2], "%x", c1);
+		else
+			return sscanf(&ref[1], "%d", c1);
+	} else { 
+		// entity reference
+		int found;
+		if ((found = LookupMnem(xml_entity_mnemonics, ref)) != -1) {
+			*c1 = found;
+			if (*c2 == 0)
+				*c2 = ' ';
+			return found;
+		}
+	}
+	return -1;
+}

src/libespeak-ng/ssml.h

                    char *outbuf,
                    int *outix,
                    int n_outbuf,
-                   bool self_closing,
                    const char *xmlbase,
                    bool *audio_text,
                    char *current_voice_id,
                    int *n_param_stack,
                    int *speech_parameters);
 
+int ParseSsmlReference(char *ref,
+                       int *c1,
+                       int *c2);
+
 #ifdef __cplusplus
 }
 #endif

src/libespeak-ng/synthesize.h

 
 void SynthesizeInit(void);
 int  Generate(PHONEME_LIST *phoneme_list, int *n_ph, bool resume);
-void MakeWave2(PHONEME_LIST *p, int n_ph);
 int  SpeakNextClause(int control);
 void SetSpeed(int control);
 void SetEmbedded(int control, int value);

src/libespeak-ng/translate.h

 //     match 1 pre 2 post 0     - use common phoneme string
 //     match 1 pre 2 post 3 0   - empty phoneme string
 
-typedef const char *constcharptr;
-
 // used to mark words with the source[] buffer
 typedef struct {
 	unsigned int flags;

tests/ssml.test

 
 test_ssml() {
 	INPUT=$1
+
+	if [ "$2" = "punct" ]
+	then
+	PARAMETERS="--punct -x"
+	else
+	PARAMETERS="-v en-US --ipa=2"
+	fi
+
 	echo "testing ${INPUT}"
 	cp $(dirname $INPUT)/$(basename ${INPUT%.*}).expected expected.txt
 	ESPEAK_DATA_PATH=`pwd` LD_LIBRARY_PATH=src:${LD_LIBRARY_PATH} \
-		src/espeak-ng -m -q -v en-US --ipa=2 -f ${INPUT} > actual.txt
+		src/espeak-ng -m -q $PARAMETERS -f ${INPUT} > actual.txt
 	diff expected.txt actual.txt || exit 1
 }
 
-for i in `ls tests/ssml/*.ssml` ; do test_ssml $i ; done
+for i in `ls tests/ssml/*.ssml` ; do test_ssml $i; done
+for i in `ls tests/ssml/*.ssml2` ; do test_ssml $i punct; done
 

tests/ssml/references.expected

+l'EsDan_: gr'eIt@D,an_: 'amp@s,and t'Ik_: kw'oUts
+b'i: b'i:
+z'Ed z'Ed
+

tests/ssml/references.ssml2

+<!-- SSML reference test
+Entity references &lt; &gt; &amp; &apos; and &quot; should be replaced in the buffer
+Character references like &#65; are translated to Unicode (65 = 'A')
+
+See https://www.tutorialspoint.com/xml/xml_syntax.htm for example
+-->
+<speak>&lt; &gt; &amp; &apos; &quot;</speak>
+<speak>B &#66;</speak>
+<speak>z &#x7A;</speak>