Update TTS service to clarify language handling and improve tests

docs/api-endpoints.md

@@ -57,13 +57,13 @@ Returns basic service metadata.
 
 ### `POST /audio/generate`
 
-Generates text-to-speech audio using Gemini or Monlam TTS (depending on language and configuration), uploads the result to S3, and returns a presigned URL.
+Generates text-to-speech audio using Monlam TTS for Tibetan (`bo`) or Gemini TTS for all other language codes, uploads the result to S3, and returns a presigned URL.
 
 **Request body:**
 
 | Field | Type | Required | Description |
 |-------|------|----------|-------------|
-| `language` | string | Yes | Language code for TTS (e.g. `bo`, `en`) |
+| `language` | string | Yes | Language code for TTS (e.g. `bo`, `en`, `fr`). Tibetan (`bo`) uses Monlam; all other codes use Gemini. |
 | `text` | string | No* | Raw text to synthesize |
 | `day_id` | UUID | No* | Plan day ID — generates audio for all text subtasks in that day |
 | `sub_task_id` | UUID | No* | Single subtask ID — generates audio for that subtask only |

tests/audio/test_tts_service.py

@@ -8,7 +8,7 @@
     _generate_gemini_tts_audio,
     _convert_to_wav,
     _parse_audio_mime_type,
-    SUPPORTED_TTS_LANGUAGES,
+    MONLAM_TTS_LANGUAGE,
 )
 from worker_api.audio.enums import PlanAudioType
 
@@ -38,10 +38,18 @@ def test_whitespace_content_raises_error(self):
         with pytest.raises(ValueError, match="Content cannot be empty"):
             generate_tts_audio("   ", PlanAudioType.RECITATION)
 
-    def test_unsupported_language_raises_error(self):
-        with pytest.raises(ValueError, match="Unsupported language for TTS"):
-            generate_tts_audio("Hello", PlanAudioType.RECITATION, language="fr")
-
+    @patch("worker_api.audio.services.tts_service._generate_gemini_tts_audio")
+    def test_other_languages_use_gemini(self, mock_gemini):
+        mock_gemini.return_value = b"fake_wav_data"
+
+        result = generate_tts_audio("Bonjour le monde", PlanAudioType.RECITATION, language="fr")
+
+        mock_gemini.assert_called_once_with(
+            content="Bonjour le monde",
+            audio_type=PlanAudioType.RECITATION,
+        )
+        assert result == b"fake_wav_data"
+
     @patch("worker_api.audio.services.tts_service.generate_monlam_tts_audio")
     def test_tibetan_language_uses_monlam(self, mock_monlam):
         mock_monlam.return_value = b"fake_audio_data"
@@ -77,9 +85,8 @@ def test_english_language_uses_gemini(self, mock_gemini):
         )
         assert result == b"fake_wav_data"
 
-    def test_supported_languages(self):
-        assert "en" in SUPPORTED_TTS_LANGUAGES
-        assert "bo" in SUPPORTED_TTS_LANGUAGES
+    def test_monlam_language_constant(self):
+        assert MONLAM_TTS_LANGUAGE == "bo"
 
 
 class TestGenerateGeminiTtsAudio:

worker_api/audio/services/tts_service.py

@@ -5,7 +5,7 @@
 from worker_api.audio.services.monlam_tts_service import generate_monlam_tts_audio
 from worker_api.audio.enums import PlanAudioType
 
-SUPPORTED_TTS_LANGUAGES = {"en", "bo"}
+MONLAM_TTS_LANGUAGE = "bo"
 
 
 def _normalize_language(language: str) -> str:
@@ -21,13 +21,7 @@ def generate_tts_audio(
     if not content.strip():
         raise ValueError("Content cannot be empty")
 
-    normalized_language = _normalize_language(language)
-    if normalized_language not in SUPPORTED_TTS_LANGUAGES:
-        raise ValueError(
-            f"Unsupported language for TTS: {language}. Supported: {', '.join(sorted(SUPPORTED_TTS_LANGUAGES))}"
-        )
-
-    if normalized_language == "bo":
+    if _normalize_language(language) == MONLAM_TTS_LANGUAGE:
         return generate_monlam_tts_audio(content, voice_name=voice_name)
 
     return _generate_gemini_tts_audio(content=content, audio_type=audio_type)