Fix non-Latin character languages falling back to English when reading Latin characters.

src/libespeak-ng/dictionary.c

@@ -2290,6 +2290,14 @@ int TranslateRules(Translator *tr, char *p_start, char *phonemes, int ph_size, c
 					if ((match1.points == 0) && ((option_sayas & 0x10) == 0)) {
 						n = utf8_in(&letter, p-1)-1;
 
+						if (tr->letter_bits_offset > 0) {
+							// not a Latin alphabet, switch to the default Latin alphabet language
+							if ((letter <= 0x241) && iswalpha(letter)) {
+								sprintf(phonemes, "%cen", phonSWITCH);
+								return 0;
+							}
+						}
+
 						// is it a bracket ?
 						if (letter == 0xe000+'(') {
 							if (pre_pause < tr->langopts.param2[LOPT_BRACKET_PAUSE])

tests/languages.test

@@ -1,11 +1,19 @@
 #!/bin/sh
 
 test_lang() {
-	TEST_LANG=$1
-	EXPECTED=$2
-	TEST_TEXT=$3
+	if test "$#" -eq 3; then
+		TEST_LANG=$1
+		TEST_NAME=$1
+		EXPECTED=$2
+		TEST_TEXT=$3
+	else
+		TEST_LANG=$1
+		TEST_NAME=$2
+		EXPECTED=$3
+		TEST_TEXT=$4
+	fi
 
-	echo "testing ${TEST_LANG}"
+	echo "testing ${TEST_NAME}"
 	ESPEAK_DATA_PATH=`pwd` LD_LIBRARY_PATH=src:${LD_LIBRARY_PATH} \
 		src/espeak-ng --stdout -v ${TEST_LANG} "${TEST_TEXT}" | \
 		sha1sum | awk '{ print $1 }' > actual.txt
@@ -30,3 +38,8 @@ test_lang en-GB-x-rp 8f1f36267f10c88ca822355ecd9d4ee37faa8228 "foot strut north
 test_lang en-US 1653355ef008d67d315aa20e630e00614472a045 "foot strut north force cure thought lot cloth trap bath"
 
 test_lang jp a3651d81da9158ee8b76bd6a31025220b4d84932 "か き く け こ"
+
+# Languages with non-Latin characters fall back to en when speaking Latin characters.
+test_lang bg "bg (English text)" ad98e816727288c7f3c525a4b4c30e72c7ea0c00 "foot strut north force cure thought lot cloth trap bath"
+test_lang fa "fa (English text)" 9460f3dcd6cc7b4f3f45a9021f7706f1bdc7ae8f "foot strut north force cure thought lot cloth trap bath"
+test_lang ka "ka (English text)" 4995fdb0fbe16bd97061c762badea03c1c44ba63 "foot strut north force cure thought lot cloth trap bath"