Add local GPU preset voices with dropdown selection.

Generate ChatTTS sample_random_speaker presets without cloud APIs; choose clone or preset in synthesize UI.

Co-authored-by: Cursor <cursoragent@cursor.com>

tts_service.py

@@ -632,12 +632,16 @@ def _concat_wavs(
     return np.concatenate(segments)
 
 
-def generate_voice(refined_text: str) -> Tuple[bool, str, Optional[str]]:
+def generate_voice(
+    refined_text: str,
+    voice_id: str = "custom",
+) -> Tuple[bool, str, Optional[str]]:
     """
-    使用 ChatTTS 将润色后的文稿合成为 wav 配音。
+    使用 ChatTTS（本地 GPU）将润色稿合成为 wav。
 
     Args:
         refined_text: LLM 润色后的配音稿
+        voice_id: ``custom`` 为锁定音色，``preset_*`` 为内置预设（见 voice_presets）
 
     Returns:
         (success, message, output_wav_path_or_none)
@@ -662,9 +666,11 @@ def generate_voice(refined_text: str) -> Tuple[bool, str, Optional[str]]:
     if chat is None:
         return False, init_err or "ChatTTS 不可用。", None
 
-    payload, spk_err = _load_speaker_payload()
+    from voice_presets import load_voice_payload
+
+    payload, spk_err = load_voice_payload(voice_id)
     if payload is None:
-        return False, spk_err or "请先锁定音色。", None
+        return False, spk_err or "请先选择或生成可用音色。", None
 
     try:
         import ChatTTS