feat(mimo): adapt MiMo-V2.5-TTS series with voicedesign and voiceclone support

astrbot/core/config/default.py

@@ -1627,6 +1627,8 @@
                         "mimo-tts-style-prompt": "",
                         "mimo-tts-dialect": "",
                         "mimo-tts-seed-text": "Hello, MiMo, have you had lunch?",
+                        "mimo-tts-user-prompt": "",
+                        "mimo-tts-voice-audio-path": "",
                         "timeout": "20",
                         "proxy": "",
                     },
@@ -2590,18 +2592,28 @@
                     "mimo-tts-style-prompt": {
                         "description": "风格提示词",
                         "type": "string",
-                        "hint": "会以 <style>...</style> 标签形式添加到待合成文本开头，用于控制语速、情绪、角色或风格，例如 开心、变快、孙悟空、悄悄话。可留空。",
+                        "hint": "用于控制语速、情绪、角色或风格，例如 开心、变快、孙悟空、悄悄话。v2 系列会以 <style>...</style> 标签拼接到文本开头，v2.5 系列会以（...）括号形式拼接。可留空。",
                     },
                     "mimo-tts-dialect": {
                         "description": "方言",
                         "type": "string",
-                        "hint": "会与风格提示词一起写入开头的 <style>...</style> 标签中，例如 东北话、四川话、河南话、粤语。可留空。",
+                        "hint": "与风格提示词一起拼接到文本开头，例如 东北话、四川话、河南话、粤语。v2 系列使用 <style> 标签，v2.5 系列使用（）括号。可留空。",
                     },
                     "mimo-tts-seed-text": {
                         "description": "种子文本",
                         "type": "string",
                         "hint": "作为可选的 user 消息发送，用于辅助调节语气和风格，不会拼接到待合成文本中。",
                     },
+                    "mimo-tts-user-prompt": {
+                        "description": "用户提示词",
+                        "type": "string",
+                        "hint": "自定义 user 角色消息。使用 voicedesign 模型时为必填项，用于通过自然语言描述音色。其他模型可留空（优先级高于种子文本）。",
+                    },
+                    "mimo-tts-voice-audio-path": {
+                        "description": "克隆音频文件路径",
+                        "type": "string",
+                        "hint": "voiceclone 模型的待模拟音频文件的本地路径，例如 /path/to/audio.wav。仅 voiceclone 模型需要填写。",
+                    },
                     "fishaudio-tts-character": {
                         "description": "character",
                         "type": "string",

astrbot/core/provider/sources/mimo_tts_api_source.py

@@ -1,5 +1,8 @@
 import base64
 import uuid
+from pathlib import Path
+
+from astrbot import logger
 
 from ..entities import ProviderType
 from ..provider import TTSProvider
@@ -41,10 +44,20 @@ def __init__(
         self.seed_text = provider_config.get(
             "mimo-tts-seed-text", DEFAULT_MIMO_TTS_SEED_TEXT
         )
+        self.user_prompt = provider_config.get("mimo-tts-user-prompt", "")
+        self.voice_audio_path = provider_config.get("mimo-tts-voice-audio-path", "")
         self.set_model(provider_config.get("model", DEFAULT_MIMO_TTS_MODEL))
         self.client = create_http_client(self.timeout, self.proxy)
 
+    def _is_v2_5(self) -> bool:
+        """Check if the current model is a v2.5 series model."""
+        return "v2.5" in self.model_name
+
     def _build_user_prompt(self) -> str | None:
+        # For voicedesign models, custom user prompt takes precedence.
+        if "voicedesign" in self.model_name and self.user_prompt.strip():
+            return self.user_prompt.strip()
+        # For other models, use seed_text as fallback.
         seed_text = self.seed_text.strip()
         return seed_text or None
 
@@ -62,13 +75,36 @@ def _build_style_prefix(self) -> str:
 
         # MiMo recommends using only the singing style tag at the very beginning.
         if "唱歌" in style_content:
+            # v2.5 uses parentheses; v2 uses <style> tags.
+            if self._is_v2_5():
+                return "（唱歌）"
             return "<style>唱歌</style>"
 
+        # v2.5 uses parentheses; v2 uses <style> tags.
+        if self._is_v2_5():
+            return f"（{style_content}）"
         return f"<style>{style_content}</style>"
 
     def _build_assistant_content(self, text: str) -> str:
         return f"{self._build_style_prefix()}{text}"
 
+    def _read_voice_audio_base64(self) -> str:
+        if not self.voice_audio_path.strip():
+            return ""
+        path = Path(self.voice_audio_path.strip())
+        if not path.exists():
+            logger.warning("Voice audio file not found: %s", path)
+            return ""
+        try:
+            suffix = path.suffix.lower().lstrip(".")
+            mime_map = {"wav": "audio/wav", "mp3": "audio/mpeg", "ogg": "audio/ogg"}
+            mime = mime_map.get(suffix, "audio/wav")
+            b64 = base64.b64encode(path.read_bytes()).decode("utf-8")
+            return f"data:{mime};base64,{b64}"
+        except Exception as exc:
+            logger.warning("Failed to read voice audio file %s: %s", path, exc)
+            return ""
+
     def _build_payload(self, text: str) -> dict:
         messages: list[dict[str, str]] = []
 
@@ -88,10 +124,16 @@ def _build_payload(self, text: str) -> dict:
             }
         )
 
-        audio_params = {"format": self.audio_format}
-        # voice design 模型不支持 audio.voice 参数
+        audio_params: dict[str, str] = {"format": self.audio_format}
         if "voicedesign" not in self.model_name:
-            audio_params["voice"] = self.voice
+            if "voiceclone" in self.model_name:
+                voice_audio_b64 = self._read_voice_audio_base64()
+                if voice_audio_b64:
+                    audio_params["voice"] = voice_audio_b64
+                else:
+                    audio_params["voice"] = self.voice
+            else:
+                audio_params["voice"] = self.voice
 
         return {
             "model": self.model_name,

dashboard/src/i18n/locales/en-US/features/config-metadata.json

@@ -1536,16 +1536,24 @@
       },
       "mimo-tts-style-prompt": {
         "description": "Style prompt",
-        "hint": "Prepended to the synthesis target text as a <style>...</style> tag to control speed, emotion, character, or style, such as happy, faster, Sun Wukong, or whispering. Optional."
+        "hint": "Controls speed, emotion, character, or style, such as happy, faster, Sun Wukong, or whispering. v2 series prepends a <style>...</style> tag; v2.5 series uses （...） parentheses. Optional."
       },
       "mimo-tts-dialect": {
         "description": "Dialect",
-        "hint": "Combined with the style prompt inside the leading <style>...</style> tag, for example Northeastern Mandarin, Sichuan dialect, Henan dialect, or Cantonese. Optional."
+        "hint": "Combined with the style prompt at the beginning of the text, for example Northeastern Mandarin, Sichuan dialect, Henan dialect, or Cantonese. v2 series uses <style> tags; v2.5 series uses （） parentheses. Optional."
       },
       "mimo-tts-seed-text": {
         "description": "Seed text",
         "hint": "Sent as an optional user message to help guide tone and speaking style. It is not appended to the synthesis target text."
       },
+      "mimo-tts-user-prompt": {
+        "description": "User prompt",
+        "hint": "Custom user role message. Required for voicedesign models to describe the voice via natural language. For other models, leave empty (takes precedence over seed text)."
+      },
+      "mimo-tts-voice-audio-path": {
+        "description": "Clone audio file path",
+        "hint": "Local path to the audio file for voiceclone models, e.g. /path/to/audio.wav. Only needed for voiceclone models."
+      },
       "fishaudio-tts-character": {
         "description": "character",
         "hint": "Fishaudio TTS character. Default is Klee. More roles: https://fish.audio/zh-CN/discovery"

dashboard/src/i18n/locales/ru-RU/features/config-metadata.json

@@ -1533,16 +1533,24 @@
             },
             "mimo-tts-style-prompt": {
                 "description": "Подсказка стиля",
-                "hint": "Добавляется в начало синтезируемого текста в виде тега <style>...</style> и управляет скоростью, эмоцией, ролью или манерой речи. Необязательно."
+                "hint": "Управляет скоростью, эмоцией, ролью или манерой речи. Серия v2 добавляет тег <style>...</style>; серия v2.5 использует скобки （...）. Необязательно."
             },
             "mimo-tts-dialect": {
                 "description": "Диалект",
-                "hint": "Объединяется с подсказкой стиля внутри начального тега <style>...</style>, например северо-восточный, сычуаньский, хэнаньский или кантонский вариант речи. Необязательно."
+                "hint": "Объединяется с подсказкой стиля в начале текста, например северо-восточный, сычуаньский, хэнаньский или кантонский. Серия v2 использует теги <style>; серия v2.5 использует скобки （）. Необязательно."
             },
             "mimo-tts-seed-text": {
                 "description": "Начальный текст",
                 "hint": "Отправляется как необязательное user-сообщение для настройки тона и манеры речи. Не добавляется к самому тексту синтеза."
             },
+            "mimo-tts-user-prompt": {
+                "description": "User-подсказка",
+                "hint": "Пользовательское user-сообщение. Обязательно для voicedesign-моделей для описания голоса на естественном языке. Для других моделей оставьте пустым (имеет приоритет над начальным текстом)."
+            },
+            "mimo-tts-voice-audio-path": {
+                "description": "Путь к аудио для клонирования",
+                "hint": "Локальный путь к аудиофайлу для voiceclone-моделей, например /path/to/audio.wav. Нужно только для voiceclone."
+            },
             "fishaudio-tts-character": {
                 "description": "Персонаж",
                 "hint": "Персонаж Fishaudio. По умолчанию Klee."

dashboard/src/i18n/locales/zh-CN/features/config-metadata.json

@@ -1538,16 +1538,24 @@
       },
       "mimo-tts-style-prompt": {
         "description": "风格提示词",
-        "hint": "会以 <style>...</style> 标签形式添加到待合成文本开头，用于控制语速、情绪、角色或风格，例如 开心、变快、孙悟空、悄悄话。可留空。"
+        "hint": "用于控制语速、情绪、角色或风格，例如 开心、变快、孙悟空、悄悄话。v2 系列会以 <style>...</style> 标签拼接到文本开头，v2.5 系列会以（...）括号形式拼接。可留空。"
       },
       "mimo-tts-dialect": {
         "description": "方言",
-        "hint": "会与风格提示词一起写入开头的 <style>...</style> 标签中，例如 东北话、四川话、河南话、粤语。可留空。"
+        "hint": "与风格提示词一起拼接到文本开头，例如 东北话、四川话、河南话、粤语。v2 系列使用 <style> 标签，v2.5 系列使用（）括号。可留空。"
       },
       "mimo-tts-seed-text": {
         "description": "种子文本",
         "hint": "作为可选的 user 消息发送，用于辅助调节语气和风格，不会拼接到待合成文本中。"
       },
+      "mimo-tts-user-prompt": {
+        "description": "用户提示词",
+        "hint": "自定义 user 角色消息。使用 voicedesign 模型时为必填项，用于通过自然语言描述音色。其他模型可留空（优先级高于种子文本）。"
+      },
+      "mimo-tts-voice-audio-path": {
+        "description": "克隆音频文件路径",
+        "hint": "voiceclone 模型的待模拟音频文件的本地路径，例如 /path/to/audio.wav。仅 voiceclone 模型需要填写。"
+      },
       "fishaudio-tts-character": {
         "description": "character",
         "hint": "fishaudio TTS 的角色。默认为可莉。更多角色请访问：https://fish.audio/zh-CN/discovery"

tests/test_mimo_api_sources.py

@@ -268,3 +268,106 @@ async def _post(*_args, **_kwargs):
         return response
 
     return _post
+
+
+def test_mimo_tts_v2_5_style_uses_parentheses():
+    provider = _make_tts_provider(
+        {
+            "model": "mimo-v2.5-tts",
+            "mimo-tts-style-prompt": "开心",
+            "mimo-tts-dialect": "四川话",
+            "mimo-tts-seed-text": "",
+        }
+    )
+    try:
+        payload = provider._build_payload("hello")
+        assert payload["messages"][0]["content"] == "（开心 四川话）hello"
+    finally:
+        asyncio.run(provider.terminate())
+
+
+def test_mimo_tts_v2_5_singing_uses_parentheses():
+    provider = _make_tts_provider(
+        {
+            "model": "mimo-v2.5-tts",
+            "mimo-tts-style-prompt": "唱歌 开心",
+            "mimo-tts-dialect": "",
+        }
+    )
+    try:
+        payload = provider._build_payload("歌词")
+        assert payload["messages"][1]["content"] == "（唱歌）歌词"
+    finally:
+        asyncio.run(provider.terminate())
+
+
+def test_mimo_tts_voicedesign_uses_custom_user_prompt():
+    provider = _make_tts_provider(
+        {
+            "model": "mimo-v2.5-tts-voicedesign",
+            "mimo-tts-user-prompt": "用活泼的声音",
+            "mimo-tts-seed-text": "",
+        }
+    )
+    try:
+        payload = provider._build_payload("hello")
+        assert payload["messages"][0] == {
+            "role": "user",
+            "content": "用活泼的声音",
+        }
+    finally:
+        asyncio.run(provider.terminate())
+
+
+def test_mimo_tts_voicedesign_seed_text_used_when_no_user_prompt():
+    provider = _make_tts_provider(
+        {
+            "model": "mimo-v2.5-tts-voicedesign",
+            "mimo-tts-user-prompt": "",
+            "mimo-tts-seed-text": "fallback seed",
+        }
+    )
+    try:
+        payload = provider._build_payload("hello")
+        assert payload["messages"][0] == {
+            "role": "user",
+            "content": "fallback seed",
+        }
+    finally:
+        asyncio.run(provider.terminate())
+
+
+def test_mimo_tts_voiceclone_uses_voice_audio(monkeypatch, tmp_path):
+    audio_file = tmp_path / "test.wav"
+    audio_file.write_bytes(b"fake audio data")
+    provider = _make_tts_provider(
+        {
+            "model": "mimo-v2.5-tts-voiceclone",
+            "mimo-tts-voice-audio-path": str(audio_file),
+            "mimo-tts-seed-text": "",
+        }
+    )
+    try:
+        payload = provider._build_payload("hello")
+        import base64
+
+        expected_b64 = base64.b64encode(b"fake audio data").decode("utf-8")
+        assert payload["audio"]["voice"] == f"data:audio/wav;base64,{expected_b64}"
+    finally:
+        asyncio.run(provider.terminate())
+
+
+def test_mimo_tts_voiceclone_falls_back_to_voice_when_no_audio():
+    provider = _make_tts_provider(
+        {
+            "model": "mimo-v2.5-tts-voiceclone",
+            "mimo-tts-voice": "preset_voice",
+            "mimo-tts-voice-audio-path": "",
+            "mimo-tts-seed-text": "",
+        }
+    )
+    try:
+        payload = provider._build_payload("hello")
+        assert payload["audio"]["voice"] == "preset_voice"
+    finally:
+        asyncio.run(provider.terminate())