change javascript and confernece to use cached speech handles

This is for better performace and to take atvantage of mrcp connections.
There should be no apparent change in usage.
The conference now supports changing the voice via the text string
by starting a text blurb with #voice# eg

#william#This will sound like William.




git-svn-id: http://svn.freeswitch.org/svn/freeswitch/trunk@5664 d0543943-73ff-0310-b7d9-9358b9ac24b2

src/mod/languages/mod_spidermonkey/mod_spidermonkey.c

@@ -1506,6 +1506,53 @@ static JSBool session_get_variable(JSContext * cx, JSObject * obj, uintN argc, j
 	return JS_TRUE;
 }
 
+static void destroy_speech_engine(struct js_session *jss)
+{
+	if (jss->speech) {
+		switch_speech_flag_t flags = SWITCH_SPEECH_FLAG_NONE;
+		switch_core_codec_destroy(&jss->speech->codec);
+		switch_core_speech_close(&jss->speech->sh, &flags);
+		jss->speech = NULL;
+	}
+}
+
+
+static switch_status_t init_speech_engine(struct js_session *jss, char *engine, char *voice)
+{
+	switch_codec_t *read_codec;
+	switch_speech_flag_t flags = SWITCH_SPEECH_FLAG_NONE;
+	uint32_t rate = 0;
+	int interval = 0;
+
+	read_codec = switch_core_session_get_read_codec(jss->session);
+	rate = read_codec->implementation->samples_per_second;
+	interval = read_codec->implementation->microseconds_per_frame / 1000;
+
+	if (switch_core_codec_init(&jss->speech->codec,
+							   "L16",
+							   NULL,
+							   rate,
+							   interval,
+							   1, SWITCH_CODEC_FLAG_ENCODE | SWITCH_CODEC_FLAG_DECODE, NULL,
+							   switch_core_session_get_pool(jss->session)) == SWITCH_STATUS_SUCCESS) {
+		switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "Raw Codec Activation Success L16@%uhz 1 channel %dms\n", rate, interval);
+	} else {
+		switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "Raw Codec Activation Failed L16@%uhz 1 channel %dms\n", rate, interval);
+		return SWITCH_STATUS_FALSE;
+	}
+
+
+	if (switch_core_speech_open(&jss->speech->sh, engine, voice, rate, interval,
+								&flags, switch_core_session_get_pool(jss->session)) != SWITCH_STATUS_SUCCESS) {
+		switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "Invalid TTS module!\n");
+		switch_core_codec_destroy(&jss->speech->codec);
+		return SWITCH_STATUS_FALSE;
+	}
+
+	return SWITCH_STATUS_SUCCESS;
+
+}
+
 
 static JSBool session_speak(JSContext * cx, JSObject * obj, uintN argc, jsval * argv, jsval * rval)
 {
@@ -1524,13 +1571,13 @@ static JSBool session_speak(JSContext * cx, JSObject * obj, uintN argc, jsval *
 
 	METHOD_SANITY_CHECK();
 
+	*rval = BOOLEAN_TO_JSVAL(JS_FALSE);
 	channel = switch_core_session_get_channel(jss->session);
 	assert(channel != NULL);
 
 	CHANNEL_SANITY_CHECK();
 
 	if (argc < 3) {
-		*rval = BOOLEAN_TO_JSVAL(JS_FALSE);
 		return JS_FALSE;
 	}
 
@@ -1538,6 +1585,34 @@ static JSBool session_speak(JSContext * cx, JSObject * obj, uintN argc, jsval *
 	voice_name = JS_GetStringBytes(JS_ValueToString(cx, argv[1]));
 	text = JS_GetStringBytes(JS_ValueToString(cx, argv[2]));
 
+
+	if (switch_strlen_zero(tts_name)) {
+		eval_some_js("~throw new Error(\"Invalid TTS Name\");", cx, obj, rval);
+		return JS_TRUE;
+	}
+
+	if (switch_strlen_zero(text)) {
+		eval_some_js("~throw new Error(\"Invalid Text\");", cx, obj, rval);
+		return JS_TRUE;
+	}
+
+
+	if (jss->speech && strcasecmp(jss->speech->sh.name, tts_name)) {
+		destroy_speech_engine(jss);
+	}
+	
+	if (jss->speech) {
+		switch_core_speech_text_param_tts(&jss->speech->sh, "voice", voice_name);
+	} else {
+		jss->speech = switch_core_session_alloc(jss->session, sizeof(*jss->speech));
+		assert(jss->speech != NULL);
+		if (init_speech_engine(jss, tts_name, voice_name) != SWITCH_STATUS_SUCCESS) {
+			eval_some_js("~throw new Error(\"Cannot allocate speech engine!\");", cx, obj, rval);
+			jss->speech = NULL;
+			return JS_TRUE;
+		}
+	}
+
 	if (argc > 3) {
 		if ((function = JS_ValueToFunction(cx, argv[3]))) {
 			memset(&cb_state, 0, sizeof(cb_state));
@@ -1555,9 +1630,6 @@ static JSBool session_speak(JSContext * cx, JSObject * obj, uintN argc, jsval *
 		}
 	}
 
-	if (!tts_name && text) {
-		return JS_FALSE;
-	}
 
 	codec = switch_core_session_get_read_codec(jss->session);
 	cb_state.ret = BOOLEAN_TO_JSVAL(JS_FALSE);
@@ -1565,8 +1637,9 @@ static JSBool session_speak(JSContext * cx, JSObject * obj, uintN argc, jsval *
 	args.input_callback = dtmf_func;
 	args.buf = bp;
 	args.buflen = len;
-	switch_ivr_speak_text(jss->session, tts_name, voice_name
-						  && strlen(voice_name) ? voice_name : NULL, codec->implementation->samples_per_second, text, &args);
+	
+	switch_core_speech_flush_tts(&jss->speech->sh);
+	switch_ivr_speak_text_handle(jss->session, &jss->speech->sh, &jss->speech->codec, NULL, text, &args);
 	JS_ResumeRequest(cx, cb_state.saveDepth);
 	*rval = cb_state.ret;
 
@@ -2389,6 +2462,8 @@ static void session_destroy(JSContext * cx, JSObject * obj)
 
 	if (cx && obj) {
 		if ((jss = JS_GetPrivate(cx, obj))) {
+			destroy_speech_engine(jss);
+			
 			if (jss->session) {
 				channel = switch_core_session_get_channel(jss->session);
 				switch_channel_set_private(channel, "jss", NULL);