fix: handle SSML phoneme attribute order in ssml_to_deepgram

src/deepgram/helpers/text_builder.py

@@ -239,12 +239,19 @@ def ssml_to_deepgram(ssml_text: str) -> str:
     # Parse XML fragments manually to handle mixed content
     # Use regex to find and replace SSML elements
 
-    # Handle <phoneme> tags
-    phoneme_pattern = r'<phoneme\s+alphabet=["\']ipa["\']\s+ph=["\'](.*?)["\']\s*>(.*?)</phoneme>'
+    # Handle <phoneme> tags. Attribute order is not significant in SSML, so match
+    # the attributes as a group and pull `ph` out of it rather than requiring
+    # `alphabet` before `ph` (which silently dropped the pronunciation otherwise).
+    phoneme_pattern = r"<phoneme\s+([^<>]*?)\s*>([^<]*)</phoneme>"
 
     def replace_phoneme(match):
-        ipa = match.group(1)
+        attributes = match.group(1)
         word = match.group(2)
+        alphabet_match = re.search(r'(?:^|\s)alphabet\s*=\s*(["\'])ipa\1(?=\s|$)', attributes)
+        ph_match = re.search(r'(?:^|\s)ph\s*=\s*(["\'])(.*?)\1(?=\s|$)', attributes)
+        if alphabet_match is None or ph_match is None:
+            return word
+        ipa = ph_match.group(2)
         return json.dumps({"word": word, "pronounce": ipa}, ensure_ascii=False)
 
     ssml_text = re.sub(phoneme_pattern, replace_phoneme, ssml_text)

tests/custom/test_text_builder.py

@@ -234,10 +234,52 @@ def test_basic_phoneme(self):
         """Test converting basic phoneme tag"""
         ssml = '<phoneme alphabet="ipa" ph="ˌæzəˈθaɪəpriːn">azathioprine</phoneme>'
         result = ssml_to_deepgram(ssml)
-        
+
         assert '"word": "azathioprine"' in result
         assert '"pronounce": "ˌæzəˈθaɪəpriːn"' in result
-
+
+    def test_phoneme_attribute_order_independent(self):
+        """ph before alphabet must work too (SSML attribute order is not significant)"""
+        ssml = '<phoneme ph="ˌæzəˈθaɪəpriːn" alphabet="ipa">azathioprine</phoneme>'
+        result = ssml_to_deepgram(ssml)
+
+        assert '"word": "azathioprine"' in result
+        assert '"pronounce": "ˌæzəˈθaɪəpriːn"' in result
+
+    def test_phoneme_attributes_allow_whitespace_around_equals(self):
+        """Valid XML whitespace around attribute equals signs must be accepted"""
+        ssml = "<phoneme ph = 'ˌæzəˈθaɪəpriːn' alphabet = \"ipa\">azathioprine</phoneme>"
+        result = ssml_to_deepgram(ssml)
+
+        assert '"word": "azathioprine"' in result
+        assert '"pronounce": "ˌæzəˈθaɪəpriːn"' in result
+
+    @pytest.mark.parametrize(
+        "attributes",
+        [
+            'ph="test"',
+            'alphabet="x-sampa" ph="test"',
+            'alphabet="ipa" data-ph="test"',
+        ],
+    )
+    def test_phoneme_requires_ipa_alphabet_and_ph_attribute(self, attributes):
+        """Unsupported or lookalike attributes must degrade to plain text"""
+        ssml = f"<phoneme {attributes}>medicine</phoneme>"
+
+        assert ssml_to_deepgram(ssml) == "medicine"
+
+    def test_unclosed_phoneme_does_not_consume_following_phoneme(self):
+        """Malformed input must not capture a later valid phoneme tag"""
+        ssml = (
+            '<phoneme alphabet="ipa" ph="bad">'
+            '<phoneme ph="good" alphabet="ipa">medicine</phoneme>'
+        )
+        result = ssml_to_deepgram(ssml)
+
+        assert '"word": "medicine"' in result
+        assert '"pronounce": "good"' in result
+        assert '"pronounce": "bad"' not in result
+
     def test_basic_break(self):
         """Test converting break tag (milliseconds)"""
         ssml = '<break time="500ms"/>'
@@ -497,4 +539,3 @@ def test_standalone_function_workflow(self):
         assert '"pronounce": "ˌæzəˈθaɪəpriːn"' in text
         assert '"word": "dupilumab"' in text
         assert '"pronounce": "duːˈpɪljuːmæb"' in text
-